Foundations and Trends

R
in
Signal Processing
Vol. 1, Nos. 1–2 (2007) 1–194
c 2007 L. R. Rabiner and R. W. Schafer
DOI: 10.1561/2000000001
Introduction to Digital Speech Processing
Lawrence R. Rabiner
1
and Ronald W. Schafer
2
1
Rutgers University and University of California, Santa Barbara, USA,
rabiner@ece.ucsb.edu
2
Hewlett-Packard Laboratories, Palo Alto, CA, USA
Abstract
Since even before the time of Alexander Graham Bell’s revolution-
ary invention, engineers and scientists have studied the phenomenon
of speech communication with an eye on creating more efficient and
effective systems of human-to-human and human-to-machine communi-
cation. Starting in the 1960s, digital signal processing (DSP), assumed
a central role in speech studies, and today DSP is the key to realizing
the fruits of the knowledge that has been gained through decades of
research. Concomitant advances in integrated circuit technology and
computer architecture have aligned to create a technological environ-
ment with virtually limitless opportunities for innovation in speech
communication applications. In this text, we highlight the central role
of DSP techniques in modern speech communication research and appli-
cations. We present a comprehensive overview of digital speech process-
ing that ranges from the basic nature of the speech signal, through a
variety of methods of representing speech in digital form, to applica-
tions in voice communication and automatic synthesis and recognition
of speech. The breadth of this subject does not allow us to discuss any
aspect of speech processing to great depth; hence our goal is to pro-
vide a useful introduction to the wide range of important concepts that
comprise the field of digital speech processing. A more comprehensive
treatment will appear in the forthcoming book, Theory and Application
of Digital Speech Processing [101].
1
Introduction
The fundamental purpose of speech is communication, i.e., the trans-
mission of messages. According to Shannon’s information theory [116],
a message represented as a sequence of discrete symbols can be quanti-
fied by its information content in bits, and the rate of transmission of
information is measured in bits/second (bps). In speech production, as
well as in many human-engineered electronic communication systems,
the information to be transmitted is encoded in the form of a contin-
uously varying (analog) waveform that can be transmitted, recorded,
manipulated, and ultimately decoded by a human listener. In the case
of speech, the fundamental analog form of the message is an acous-
tic waveform, which we call the speech signal. Speech signals, as illus-
trated in Figure 1.1, can be converted to an electrical waveform by
a microphone, further manipulated by both analog and digital signal
processing, and then converted back to acoustic form by a loudspeaker,
a telephone handset or headphone, as desired. This form of speech pro-
cessing is, of course, the basis for Bell’s telephone invention as well as
today’s multitude of devices for recording, transmitting, and manip-
ulating speech and audio signals. Although Bell made his invention
without knowing the fundamentals of information theory, these ideas
3
4 Introduction
0 0.02 0.04 0.06 0.08 0.1 0.12
0.48
0.36
0.24
0.12
0
time in seconds
SH
UH
D W IY
CH
EY
S
Fig. 1.1 A speech waveform with phonetic labels for the text message “Should we chase.”
have assumed great importance in the design of sophisticated modern
communications systems. Therefore, even though our main focus will
be mostly on the speech waveform and its representation in the form of
parametric models, it is nevertheless useful to begin with a discussion
of how information is encoded in the speech waveform.
1.1 The Speech Chain
Figure 1.2 shows the complete process of producing and perceiving
speech from the formulation of a message in the brain of a talker, to
the creation of the speech signal, and finally to the understanding of
the message by a listener. In their classic introduction to speech sci-
ence, Denes and Pinson aptly referred to this process as the “speech
chain” [29]. The process starts in the upper left as a message repre-
sented somehow in the brain of the speaker. The message information
can be thought of as having a number of different representations dur-
ing the process of speech production (the upper path in Figure 1.2).
1.1 The Speech Chain 5
Message
Formulation
Message
Understanding
Language
Translation
Language
Code
Neuro-Muscular
Controls
Vocal Tract
System
Transmission
Channel
acoustic
waveform
acoustic
waveform
Neural
Transduction
Basilar
Membrane
Motion
discrete input continuous input
Speech Production
Speech Perception
continuous output discrete output
text
phonemes,
prosody
articulatory motions
semantics
phonemes, words
sentences
feature
extraction
spectrum
analysis
50 bps 200 bps 2000 bps 64-700 Kbps
information rate
excitation,
formants
Fig. 1.2 The Speech Chain: from message, to speech signal, to understanding.
For example the message could be represented initially as English text.
In order to “speak” the message, the talker implicitly converts the text
into a symbolic representation of the sequence of sounds corresponding
to the spoken version of the text. This step, called the language code
generator in Figure 1.2, converts text symbols to phonetic symbols
(along with stress and durational information) that describe the basic
sounds of a spoken version of the message and the manner (i.e., the
speed and emphasis) in which the sounds are intended to be produced.
As an example, the segments of the waveform of Figure 1.1 are labeled
with phonetic symbols using a computer-keyboard-friendly code called
ARPAbet.
1
Thus, the text “should we chase” is represented phoneti-
cally (in ARPAbet symbols) as [SH UH D — W IY — CH EY S]. (See
Chapter 2 for more discussion of phonetic transcription.) The third step
in the speech production process is the conversion to “neuro-muscular
controls,” i.e., the set of control signals that direct the neuro-muscular
system to move the speech articulators, namely the tongue, lips, teeth,
1
The International Phonetic Association (IPA) provides a set of rules for phonetic tran-
scription using an equivalent set of specialized symbols. The ARPAbet code does not
require special fonts and is thus more convenient for computer applications.
6 Introduction
jaw and velum, in a manner that is consistent with the sounds of the
desired spoken message and with the desired degree of emphasis. The
end result of the neuro-muscular controls step is a set of articulatory
motions (continuous control) that cause the vocal tract articulators to
move in a prescribed manner in order to create the desired sounds.
Finally the last step in the Speech Production process is the “vocal
tract system” that physically creates the necessary sound sources and
the appropriate vocal tract shapes over time so as to create an acoustic
waveform, such as the one shown in Figure 1.1, that encodes the infor-
mation in the desired message into the speech signal.
To determine the rate of information flow during speech produc-
tion, assume that there are about 32 symbols (letters) in the language
(in English there are 26 letters, but if we include simple punctuation
we get a count closer to 32 = 2
5
symbols). Furthermore, the rate of
speaking for most people is about 10 symbols per second (somewhat
on the high side, but still acceptable for a rough information rate esti-
mate). Hence, assuming independent letters as a simple approximation,
we estimate the base information rate of the text message as about
50 bps (5 bits per symbol times 10 symbols per second). At the second
stage of the process, where the text representation is converted into
phonemes and prosody (e.g., pitch and stress) markers, the informa-
tion rate is estimated to increase by a factor of 4 to about 200 bps. For
example, the ARBAbet phonetic symbol set used to label the speech
sounds in Figure 1.1 contains approximately 64 = 2
6
symbols, or about
6 bits/phoneme (again a rough approximation assuming independence
of phonemes). In Figure 1.1, there are 8 phonemes in approximately
600 ms. This leads to an estimate of 8 6/0.6 = 80 bps. Additional
information required to describe prosodic features of the signal (e.g.,
duration, pitch, loudness) could easily add 100 bps to the total infor-
mation rate for a message encoded as a speech signal.
The information representations for the first two stages in the speech
chain are discrete so we can readily estimate the rate of information
flow with some simple assumptions. For the next stage in the speech
production part of the speech chain, the representation becomes con-
tinuous (in the form of control signals for articulatory motion). If they
could be measured, we could estimate the spectral bandwidth of these
1.1 The Speech Chain 7
control signals and appropriately sample and quantize these signals to
obtain equivalent digital signals for which the data rate could be esti-
mated. The articulators move relatively slowly compared to the time
variation of the resulting acoustic waveform. Estimates of bandwidth
and required accuracy suggest that the total data rate of the sampled
articulatory control signals is about 2000 bps [34]. Thus, the original
text message is represented by a set of continuously varying signals
whose digital representation requires a much higher data rate than
the information rate that we estimated for transmission of the mes-
sage as a speech signal.
2
Finally, as we will see later, the data rate
of the digitized speech waveform at the end of the speech production
part of the speech chain can be anywhere from 64,000 to more than
700,000 bps. We arrive at such numbers by examining the sampling
rate and quantization required to represent the speech signal with a
desired perceptual fidelity. For example, “telephone quality” requires
that a bandwidth of 0–4 kHz be preserved, implying a sampling rate of
8000 samples/s. Each sample can be quantized with 8 bits on a log scale,
resulting in a bit rate of 64,000 bps. This representation is highly intelli-
gible (i.e., humans can readily extract the message from it) but to most
listeners, it will sound different from the original speech signal uttered
by the talker. On the other hand, the speech waveform can be repre-
sented with “CD quality” using a sampling rate of 44,100 samples/s
with 16 bit samples, or a data rate of 705,600 bps. In this case, the
reproduced acoustic signal will be virtually indistinguishable from the
original speech signal.
As we move from text to speech waveform through the speech chain,
the result is an encoding of the message that can be effectively transmit-
ted by acoustic wave propagation and robustly decoded by the hear-
ing mechanism of a listener. The above analysis of data rates shows
that as we move from text to sampled speech waveform, the data rate
can increase by a factor of 10,000. Part of this extra information rep-
resents characteristics of the talker such as emotional state, speech
mannerisms, accent, etc., but much of it is due to the inefficiency
2
Note that we introduce the term data rate for digital representations to distinguish from
the inherent information content of the message represented by the speech signal.
8 Introduction
of simply sampling and finely quantizing analog signals. Thus, moti-
vated by an awareness of the low intrinsic information rate of speech,
a central theme of much of digital speech processing is to obtain a
digital representation with lower data rate than that of the sampled
waveform.
The complete speech chain consists of a speech production/
generation model, of the type discussed above, as well as a speech
perception/recognition model, as shown progressing to the left in the
bottom half of Figure 1.2. The speech perception model shows the series
of steps from capturing speech at the ear to understanding the mes-
sage encoded in the speech signal. The first step is the effective con-
version of the acoustic waveform to a spectral representation. This is
done within the inner ear by the basilar membrane, which acts as a
non-uniform spectrum analyzer by spatially separating the spectral
components of the incoming speech signal and thereby analyzing them
by what amounts to a non-uniform filter bank. The next step in the
speech perception process is a neural transduction of the spectral fea-
tures into a set of sound features (or distinctive features as they are
referred to in the area of linguistics) that can be decoded and processed
by the brain. The next step in the process is a conversion of the sound
features into the set of phonemes, words, and sentences associated with
the in-coming message by a language translation process in the human
brain. Finally, the last step in the speech perception model is the con-
version of the phonemes, words and sentences of the message into an
understanding of the meaning of the basic message in order to be able
to respond to or take some appropriate action. Our fundamental under-
standing of the processes in most of the speech perception modules in
Figure 1.2 is rudimentary at best, but it is generally agreed that some
physical correlate of each of the steps in the speech perception model
occur within the human brain, and thus the entire model is useful for
thinking about the processes that occur.
There is one additional process shown in the diagram of the com-
plete speech chain in Figure 1.2 that we have not discussed — namely
the transmission channel between the speech generation and speech
perception parts of the model. In its simplest embodiment, this trans-
mission channel consists of just the acoustic wave connection between
1.2 Applications of Digital Speech Processing 9
a speaker and a listener who are in a common space. It is essen-
tial to include this transmission channel in our model for the speech
chain since it includes real world noise and channel distortions that
make speech and message understanding more difficult in real com-
munication environments. More interestingly for our purpose here —
it is in this domain that we find the applications of digital speech
processing.
1.2 Applications of Digital Speech Processing
The first step in most applications of digital speech processing is to
convert the acoustic waveform to a sequence of numbers. Most modern
A-to-D converters operate by sampling at a very high rate, applying a
digital lowpass filter with cutoff set to preserve a prescribed bandwidth,
and then reducing the sampling rate to the desired sampling rate, which
can be as low as twice the cutoff frequency of the sharp-cutoff digital
filter. This discrete-time representation is the starting point for most
applications. From this point, other representations are obtained by
digital processing. For the most part, these alternative representations
are based on incorporating knowledge about the workings of the speech
chain as depicted in Figure 1.2. As we will see, it is possible to incor-
porate aspects of both the speech production and speech perception
process into the digital representation and processing. It is not an over-
simplification to assert that digital speech processing is grounded in a
set of techniques that have the goal of pushing the data rate of the
speech representation to the left along either the upper or lower path
in Figure 1.2.
The remainder of this chapter is devoted to a brief summary of the
applications of digital speech processing, i.e., the systems that people
interact with daily. Our discussion will confirm the importance of the
digital representation in all application areas.
1.2.1 Speech Coding
Perhaps the most widespread applications of digital speech process-
ing technology occur in the areas of digital transmission and storage
10 Introduction
A-to-D
Converter
Analysis/
Encoding
Channel or
Medium
Synthesis/
Decoding
D-to-A
Converter
speech
signal
samples
data
samples data
) (t x
c
] [n x ] [n y
] [n y ] [n x ) (t x
c
decoded
signal
Fig. 1.3 Speech coding block diagram — encoder and decoder.
of speech signals. In these areas the centrality of the digital repre-
sentation is obvious, since the goal is to compress the digital wave-
form representation of speech into a lower bit-rate representation. It
is common to refer to this activity as “speech coding” or “speech
compression.”
Figure 1.3 shows a block diagram of a generic speech encod-
ing/decoding (or compression) system. In the upper part of the figure,
the A-to-D converter converts the analog speech signal x
c
(t) to a sam-
pled waveform representation x[n]. The digital signal x[n] is analyzed
and coded by digital computation algorithms to produce a new digital
signal y[n] that can be transmitted over a digital communication chan-
nel or stored in a digital storage medium as ˆ y[n]. As we will see, there
are a myriad of ways to do the encoding so as to reduce the data rate
over that of the sampled and quantized speech waveform x[n]. Because
the digital representation at this point is often not directly related to
the sampled speech waveform, y[n] and ˆ y[n] are appropriately referred
to as data signals that represent the speech signal. The lower path in
Figure 1.3 shows the decoder associated with the speech coder. The
received data signal ˆ y[n] is decoded using the inverse of the analysis
processing, giving the sequence of samples ˆ x[n] which is then converted
(using a D-to-A Converter) back to an analog signal ˆ x
c
(t) for human
listening. The decoder is often called a synthesizer because it must
reconstitute the speech waveform from data that may bear no direct
relationship to the waveform.
1.2 Applications of Digital Speech Processing 11
With carefully designed error protection coding of the digital
representation, the transmitted (y[n]) and received (ˆ y[n]) data can be
essentially identical. This is the quintessential feature of digital coding.
In theory, perfect transmission of the coded digital representation is
possible even under very noisy channel conditions, and in the case of
digital storage, it is possible to store a perfect copy of the digital repre-
sentation in perpetuity if sufficient care is taken to update the storage
medium as storage technology advances. This means that the speech
signal can be reconstructed to within the accuracy of the original cod-
ing for as long as the digital representation is retained. In either case,
the goal of the speech coder is to start with samples of the speech signal
and reduce (compress) the data rate required to represent the speech
signal while maintaining a desired perceptual fidelity. The compressed
representation can be more efficiently transmitted or stored, or the bits
saved can be devoted to error protection.
Speech coders enable a broad range of applications including nar-
rowband and broadband wired telephony, cellular communications,
voice over internet protocol (VoIP) (which utilizes the internet as
a real-time communications medium), secure voice for privacy and
encryption (for national security applications), extremely narrowband
communications channels (such as battlefield applications using high
frequency (HF) radio), and for storage of speech for telephone answer-
ing machines, interactive voice response (IVR) systems, and pre-
recorded messages. Speech coders often utilize many aspects of both
the speech production and speech perception processes, and hence may
not be useful for more general audio signals such as music. Coders that
are based on incorporating only aspects of sound perception generally
do not achieve as much compression as those based on speech produc-
tion, but they are more general and can be used for all types of audio
signals. These coders are widely deployed in MP3 and AAC players and
for audio in digital television systems [120].
1.2.2 Text-to-Speech Synthesis
For many years, scientists and engineers have studied the speech pro-
duction process with the goal of building a system that can start with
12 Introduction
Linguistic
Rules
Synthesis
Algorithm
D-to-A
Converter
speech
text
Fig. 1.4 Text-to-speech synthesis system block diagram.
text and produce speech automatically. In a sense, a text-to-speech
synthesizer such as depicted in Figure 1.4 is a digital simulation of the
entire upper part of the speech chain diagram. The input to the system
is ordinary text such as an email message or an article from a newspa-
per or magazine. The first block in the text-to-speech synthesis system,
labeled linguistic rules, has the job of converting the printed text input
into a set of sounds that the machine must synthesize. The conversion
from text to sounds involves a set of linguistic rules that must determine
the appropriate set of sounds (perhaps including things like emphasis,
pauses, rates of speaking, etc.) so that the resulting synthetic speech
will express the words and intent of the text message in what passes
for a natural voice that can be decoded accurately by human speech
perception. This is more difficult than simply looking up the words in
a pronouncing dictionary because the linguistic rules must determine
how to pronounce acronyms, how to pronounce ambiguous words like
read, bass, object, how to pronounce abbreviations like St. (street or
Saint), Dr. (Doctor or drive), and how to properly pronounce proper
names, specialized terms, etc. Once the proper pronunciation of the
text has been determined, the role of the synthesis algorithm is to cre-
ate the appropriate sound sequence to represent the text message in
the form of speech. In essence, the synthesis algorithm must simulate
the action of the vocal tract system in creating the sounds of speech.
There are many procedures for assembling the speech sounds and com-
piling them into a proper sentence, but the most promising one today is
called “unit selection and concatenation.” In this method, the computer
stores multiple versions of each of the basic units of speech (phones, half
phones, syllables, etc.), and then decides which sequence of speech units
sounds best for the particular text message that is being produced. The
basic digital representation is not generally the sampled speech wave.
Instead, some sort of compressed representation is normally used to
1.2 Applications of Digital Speech Processing 13
save memory and, more importantly, to allow convenient manipulation
of durations and blending of adjacent sounds. Thus, the speech syn-
thesis algorithm would include an appropriate decoder, as discussed in
Section 1.2.1, whose output is converted to an analog representation
via the D-to-A converter.
Text-to-speech synthesis systems are an essential component of
modern human–machine communications systems and are used to do
things like read email messages over a telephone, provide voice out-
put from GPS systems in automobiles, provide the voices for talking
agents for completion of transactions over the internet, handle call cen-
ter help desks and customer care applications, serve as the voice for
providing information from handheld devices such as foreign language
phrasebooks, dictionaries, crossword puzzle helpers, and as the voice of
announcement machines that provide information such as stock quotes,
airline schedules, updates on arrivals and departures of flights, etc.
Another important application is in reading machines for the blind,
where an optical character recognition system provides the text input
to a speech synthesis system.
1.2.3 Speech Recognition and Other Pattern
Matching Problems
Another large class of digital speech processing applications is con-
cerned with the automatic extraction of information from the speech
signal. Most such systems involve some sort of pattern matching.
Figure 1.5 shows a block diagram of a generic approach to pattern
matching problems in speech processing. Such problems include the
following: speech recognition, where the object is to extract the mes-
sage from the speech signal; speaker recognition, where the goal is
to identify who is speaking; speaker verification, where the goal is
to verify a speaker’s claimed identity from analysis of their speech
A-to-D
Converter
Feature
Analysis
Pattern
Matching
symbols
speech
Fig. 1.5 Block diagram of general pattern matching system for speech signals.
14 Introduction
signal; word spotting, which involves monitoring a speech signal for
the occurrence of specified words or phrases; and automatic index-
ing of speech recordings based on recognition (or spotting) of spoken
keywords.
The first block in the pattern matching system converts the ana-
log speech waveform to digital form using an A-to-D converter. The
feature analysis module converts the sampled speech signal to a set
of feature vectors. Often, the same analysis techniques that are used
in speech coding are also used to derive the feature vectors. The final
block in the system, namely the pattern matching block, dynamically
time aligns the set of feature vectors representing the speech signal with
a concatenated set of stored patterns, and chooses the identity associ-
ated with the pattern which is the closest match to the time-aligned set
of feature vectors of the speech signal. The symbolic output consists
of a set of recognized words, in the case of speech recognition, or the
identity of the best matching talker, in the case of speaker recognition,
or a decision as to whether to accept or reject the identity claim of a
speaker in the case of speaker verification.
Although the block diagram of Figure 1.5 represents a wide range
of speech pattern matching problems, the biggest use has been in the
area of recognition and understanding of speech in support of human–
machine communication by voice. The major areas where such a system
finds applications include command and control of computer software,
voice dictation to create letters, memos, and other documents, natu-
ral language voice dialogues with machines to enable help desks and
call centers, and for agent services such as calendar entry and update,
address list modification and entry, etc.
Pattern recognition applications often occur in conjunction with
other digital speech processing applications. For example, one of the
preeminent uses of speech technology is in portable communication
devices. Speech coding at bit rates on the order of 8 Kbps enables nor-
mal voice conversations in cell phones. Spoken name speech recognition
in cellphones enables voice dialing capability that can automatically
dial the number associated with the recognized name. Names from
directories with upwards of several hundred names can readily be rec-
ognized and dialed using simple speech recognition technology.
1.2 Applications of Digital Speech Processing 15
Another major speech application that has long been a dream
of speech researchers is automatic language translation. The goal of
language translation systems is to convert spoken words in one lan-
guage to spoken words in another language so as to facilitate natural
language voice dialogues between people speaking different languages.
Language translation technology requires speech synthesis systems that
work in both languages, along with speech recognition (and generally
natural language understanding) that also works for both languages;
hence it is a very difficult task and one for which only limited progress
has been made. When such systems exist, it will be possible for people
speaking different languages to communicate at data rates on the order
of that of printed text reading!
1.2.4 Other Speech Applications
The range of speech communication applications is illustrated in
Figure 1.6. As seen in this figure, the techniques of digital speech
processing are a key ingredient of a wide range of applications that
include the three areas of transmission/storage, speech synthesis, and
speech recognition as well as many others such as speaker identification,
speech signal quality enhancement, and aids for the hearing- or visually-
impaired.
The block diagram in Figure 1.7 represents any system where time
signals such as speech are processed by the techniques of DSP. This
figure simply depicts the notion that once the speech signal is sampled,
it can be manipulated in virtually limitless ways by DSP techniques.
Here again, manipulations and modifications of the speech signal are
Digital
Transmission
& Storage
Digital Speech Processing
Techniques
Speech
Synthesis
Speech
Recognition
Speaker
Verification/
Identification
Enhancement
of Speech
Quality
Aids for the
Handicapped
Fig. 1.6 Range of speech communication applications.
16 Introduction
A-to-D
Converter
Computer
Algorithm
D-to-A
Converter
speech
speech
Fig. 1.7 General block diagram for application of digital signal processing to speech signals.
usually achieved by transforming the speech signal into an alternative
representation (that is motivated by our understanding of speech pro-
duction and speech perception), operating on that representation by
further digital computation, and then transforming back to the wave-
form domain, using a D-to-A converter.
One important application area is speech enhancement, where the
goal is to remove or suppress noise or echo or reverberation picked up by
a microphone along with the desired speech signal. In human-to-human
communication, the goal of speech enhancement systems is to make the
speech more intelligible and more natural; however, in reality the best
that has been achieved so far is less perceptually annoying speech that
essentially maintains, but does not improve, the intelligibility of the
noisy speech. Success has been achieved, however, in making distorted
speech signals more useful for further processing as part of a speech
coder, synthesizer, or recognizer. An excellent reference in this area is
the recent textbook by Loizou [72].
Other examples of manipulation of the speech signal include
timescale modification to align voices with video segments, to mod-
ify voice qualities, and to speed-up or slow-down prerecorded speech
(e.g., for talking books, rapid review of voice mail messages, or careful
scrutinizing of spoken material).
1.3 Our Goal for this Text
We have discussed the speech signal and how it encodes information
for human communication. We have given a brief overview of the way
in which digital speech processing is being applied today, and we have
hinted at some of the possibilities that exist for the future. These and
many more examples all rely on the basic principles of digital speech
processing, which we will discuss in the remainder of this text. We
make no pretense of exhaustive coverage. The subject is too broad and
1.3 Our Goal for this Text 17
too deep. Our goal is only to provide an up-to-date introduction to this
fascinating field. We will not be able to go into great depth, and we will
not be able to cover all the possible applications of digital speech pro-
cessing techniques. Instead our focus is on the fundamentals of digital
speech processing and their application to coding, synthesis, and recog-
nition. This means that some of the latest algorithmic innovations and
applications will not be discussed — not because they are not interest-
ing, but simply because there are so many fundamental tried-and-true
techniques that remain at the core of digital speech processing. We
hope that this text will stimulate readers to investigate the subject in
greater depth using the extensive set of references provided.
2
The Speech Signal
As the discussion in Chapter 1 shows, the goal in many applications of
digital speech processing techniques is to move the digital representa-
tion of the speech signal from the waveform samples back up the speech
chain toward the message. To gain a better idea of what this means,
this chapter provides a brief overview of the phonetic representation of
speech and an introduction to models for the production of the speech
signal.
2.1 Phonetic Representation of Speech
Speech can be represented phonetically by a finite set of symbols called
the phonemes of the language, the number of which depends upon the
language and the refinement of the analysis. For most languages the
number of phonemes is between 32 and 64. A condensed inventory of
the sounds of speech in the English language is given in Table 2.1,
where the phonemes are denoted by a set of ASCII symbols called the
ARPAbet. Table 2.1 also includes some simple examples of ARPAbet
transcriptions of words containing each of the phonemes of English.
Additional phonemes can be added to Table 2.1 to account for allo-
phonic variations and events such as glottal stops and pauses.
18
2.1 Phonetic Representation of Speech 19
Table 2.1 Condensed list of ARPAbet phonetic symbols for North American English.
Class ARPAbet Example Transcription
Vowels and IY beet [B IY T]
diphthongs IH bit [B IH T]
EY bait [B EY T]
EH bet [B EH T]
AE bat [B AE T]
AA bob [B AA B]
AO born [B AO R N]
UH book [B UH K]
OW boat [B OW T]
UW boot [B UW T]
AH but [B AH T ]
ER bird [B ER D]
AY buy [B AY]
AW down [D AW N]
OY boy [B OY]
Glides Y you [Y UH]
R rent [R EH N T]
Liquids W wit [W IH T]
L l et [L EH T]
Nasals M met [M EH T]
N net [N EH T]
NG sing [S IH NG]
Stops P pat [P AE T]
B bet [B EH T]
T ten [T EH N]
D debt [D EH T]
K kit [K IH T]
G get [G EH T]
Fricatives HH hat [HH AE T]
F f at [F AE T]
V vat [V AE T]
TH thing [TH IH NG]
DH that [DH AE T]
S sat [S AE T]
Z zoo [Z UW]
SH shut [SH AH T]
ZH azure [AE ZH ER]
Affricates CH chase [CH EY S]
JH j udge [JH AH JH ]
a
This set of 39 phonemes is used in the CMU Pronouncing Dictionary available on-line at
http://www.speech.cs.cmu.edu/cgi-bin/cmudict.
Figure 1.1 on p. 4 shows how the sounds corresponding to the text
“should we chase” are encoded into a speech waveform. We see that, for
the most part, phonemes have a distinctive appearance in the speech
waveform. Thus sounds like /SH/ and /S/ look like (spectrally shaped)
20 The Speech Signal
random noise, while the vowel sounds /UH/, /IY/, and /EY/ are highly
structured and quasi-periodic. These differences result from the distinc-
tively different ways that these sounds are produced.
2.2 Models for Speech Production
A schematic longitudinal cross-sectional drawing of the human vocal
tract mechanism is given in Figure 2.1 [35]. This diagram highlights
the essential physical features of human anatomy that enter into the
final stages of the speech production process. It shows the vocal tract
as a tube of nonuniform cross-sectional area that is bounded at one end
by the vocal cords and at the other by the mouth opening. This tube
serves as an acoustic transmission system for sounds generated inside
the vocal tract. For creating nasal sounds like /M/, /N/, or /NG/,
a side-branch tube, called the nasal tract, is connected to the main
acoustic branch by the trapdoor action of the velum. This branch path
radiates sound at the nostrils. The shape (variation of cross-section
along the axis) of the vocal tract varies with time due to motions of
the lips, jaw, tongue, and velum. Although the actual human vocal
tract is not laid out along a straight line as in Figure 2.1, this type of
model is a reasonable approximation for wavelengths of the sounds in
speech.
The sounds of speech are generated in the system of Figure 2.1 in
several ways. Voiced sounds (vowels, liquids, glides, nasals in Table 2.1)
Fig. 2.1 Schematic model of the vocal tract system. (After Flanagan et al. [35].)
2.2 Models for Speech Production 21
are produced when the vocal tract tube is excited by pulses of air pres-
sure resulting from quasi-periodic opening and closing of the glottal
orifice (opening between the vocal cords). Examples in Figure 1.1 are
the vowels /UH/, /IY/, and /EY/, and the liquid consonant /W/.
Unvoiced sounds are produced by creating a constriction somewhere in
the vocal tract tube and forcing air through that constriction, thereby
creating turbulent air flow, which acts as a random noise excitation of
the vocal tract tube. Examples are the unvoiced fricative sounds such as
/SH/ and /S/. A third sound production mechanism is when the vocal
tract is partially closed off causing turbulent flow due to the constric-
tion, at the same time allowing quasi-periodic flow due to vocal cord
vibrations. Sounds produced in this manner include the voiced frica-
tives /V/, /DH/, /Z/, and /ZH/. Finally, plosive sounds such as /P/,
/T/, and /K/ and affricates such as /CH/ are formed by momentarily
closing off air flow, allowing pressure to build up behind the closure, and
then abruptly releasing the pressure. All these excitation sources cre-
ate a wide-band excitation signal to the vocal tract tube, which acts as
an acoustic transmission line with certain vocal tract shape-dependent
resonances that tend to emphasize some frequencies of the excitation
relative to others.
As discussed in Chapter 1 and illustrated by the waveform in
Figure 1.1, the general character of the speech signal varies at the
phoneme rate, which is on the order of 10 phonemes per second, while
the detailed time variations of the speech waveform are at a much higher
rate. That is, the changes in vocal tract configuration occur relatively
slowly compared to the detailed time variation of the speech signal. The
sounds created in the vocal tract are shaped in the frequency domain
by the frequency response of the vocal tract. The resonance frequencies
resulting from a particular configuration of the articulators are instru-
mental in forming the sound corresponding to a given phoneme. These
resonance frequencies are called the formant frequencies of the sound
[32, 34]. In summary, the fine structure of the time waveform is created
by the sound sources in the vocal tract, and the resonances of the vocal
tract tube shape these sound sources into the phonemes.
The system of Figure 2.1 can be described by acoustic theory,
and numerical techniques can be used to create a complete physical
22 The Speech Signal
Linear
System
Excitation
Generator
excitation signal speech signal
Vocal Tract
Parameters
Excitation
Parameters
Fig. 2.2 Source/system model for a speech signal.
simulation of sound generation and transmission in the vocal tract
[36, 93], but, for the most part, it is sufficient to model the produc-
tion of a sampled speech signal by a discrete-time system model such
as the one depicted in Figure 2.2. The discrete-time time-varying linear
system on the right in Figure 2.2 simulates the frequency shaping of
the vocal tract tube. The excitation generator on the left simulates the
different modes of sound generation in the vocal tract. Samples of a
speech signal are assumed to be the output of the time-varying linear
system.
In general such a model is called a source/system model of speech
production. The short-time frequency response of the linear system
simulates the frequency shaping of the vocal tract system, and since the
vocal tract changes shape relatively slowly, it is reasonable to assume
that the linear system response does not vary over time intervals on the
order of 10 ms or so. Thus, it is common to characterize the discrete-
time linear system by a system function of the form:
H(z) =
M
¸
k=0
b
k
z
−k
1 −
N
¸
k=1
a
k
z
−k
=
b
0
M
¸
k=1
(1 − d
k
z
−1
)
N
¸
k=1
(1 − c
k
z
−1
)
, (2.1)
where the filter coefficients a
k
and b
k
(labeled as vocal tract parameters
in Figure 2.2) change at a rate on the order of 50–100 times/s. Some
of the poles (c
k
) of the system function lie close to the unit circle
and create resonances to model the formant frequencies. In detailed
modeling of speech production [32, 34, 64], it is sometimes useful to
employ zeros (d
k
) of the system function to model nasal and fricative
2.2 Models for Speech Production 23
sounds. However, as we discuss further in Chapter 4, many applications
of the source/system model only include poles in the model because this
simplifies the analysis required to estimate the parameters of the model
from the speech signal.
The box labeled excitation generator in Figure 2.2 creates an appro-
priate excitation for the type of sound being produced. For voiced
speech the excitation to the linear system is a quasi-periodic sequence
of discrete (glottal) pulses that look very much like those shown in the
righthand half of the excitation signal waveform in Figure 2.2. The fun-
damental frequency of the glottal excitation determines the perceived
pitch of the voice. The individual finite-duration glottal pulses have a
lowpass spectrum that depends on a number of factors [105]. There-
fore, the periodic sequence of smooth glottal pulses has a harmonic line
spectrum with components that decrease in amplitude with increasing
frequency. Often it is convenient to absorb the glottal pulse spectrum
contribution into the vocal tract system model of (2.1). This can be
achieved by a small increase in the order of the denominator over what
would be needed to represent the formant resonances. For unvoiced
speech, the linear system is excited by a random number generator
that produces a discrete-time noise signal with flat spectrum as shown
in the left-hand half of the excitation signal. The excitation in Fig-
ure 2.2 switches from unvoiced to voiced leading to the speech signal
output as shown in the figure. In either case, the linear system imposes
its frequency response on the spectrum to create the speech sounds.
This model of speech as the output of a slowly time-varying digital
filter with an excitation that captures the nature of the voiced/unvoiced
distinction in speech production is the basis for thinking about the
speech signal, and a wide variety of digital representations of the speech
signal are based on it. That is, the speech signal is represented by the
parameters of the model instead of the sampled waveform. By assuming
that the properties of the speech signal (and the model) are constant
over short time intervals, it is possible to compute/measure/estimate
the parameters of the model by analyzing short blocks of samples of
the speech signal. It is through such models and analysis techniques
that we are able to build properties of the speech production process
into digital representations of the speech signal.
24 The Speech Signal
2.3 More Refined Models
Source/system models as shown in Figure 2.2 with the system char-
acterized by a time-sequence of time-invariant systems are quite suffi-
cient for most applications in speech processing, and we shall rely on
such models throughout this text. However, such models are based on
many approximations including the assumption that the source and the
system do not interact, the assumption of linearity, and the assump-
tion that the distributed continuous-time vocal tract transmission sys-
tem can be modeled by a discrete linear time-invariant system. Fluid
mechanics and acoustic wave propagation theory are fundamental phys-
ical principles that must be applied for detailed modeling of speech
production. Since the early work of Flanagan and Ishizaka [34, 36, 51]
much work has been devoted to creating detailed simulations of glottal
flow, the interaction of the glottal source and the vocal tract in speech
production, and the nonlinearities that enter into sound generation and
transmission in the vocal tract. Stevens [121] and Quatieri [94] provide
useful discussions of these effects. For many years, researchers have
sought to measure the physical dimensions of the human vocal tract
during speech production. This information is essential for detailed sim-
ulations based on acoustic theory. Early efforts to measure vocal tract
area functions involved hand tracing on X-ray pictures [32]. Recent
advances in MRI imaging and computer image analysis have provided
significant advances in this area of speech science [17].
3
Hearing and Auditory Perception
In Chapter 2, we introduced the speech production process and showed
how we could model speech production using discrete-time systems.
In this chapter we turn to the perception side of the speech chain to
discuss properties of human sound perception that can be employed to
create digital representations of the speech signal that are perceptually
robust.
3.1 The Human Ear
Figure 3.1 shows a schematic view of the human ear showing the three
distinct sound processing sections, namely: the outer ear consisting of
the pinna, which gathers sound and conducts it through the external
canal to the middle ear; the middle ear beginning at the tympanic
membrane, or eardrum, and including three small bones, the malleus
(also called the hammer), the incus (also called the anvil) and the stapes
(also called the stirrup), which perform a transduction from acoustic
waves to mechanical pressure waves; and finally, the inner ear, which
consists of the cochlea and the set of neural connections to the auditory
nerve, which conducts the neural signals to the brain.
25
26 Hearing and Auditory Perception
Fig. 3.1 Schematic view of the human ear (inner and middle structures enlarged). (After
Flanagan [34].)
Figure 3.2 [107] depicts a block diagram abstraction of the auditory
system. The acoustic wave is transmitted from the outer ear to the
inner ear where the ear drum and bone structures convert the sound
wave to mechanical vibrations which ultimately are transferred to the
basilar membrane inside the cochlea. The basilar membrane vibrates in
a frequency-selective manner along its extent and thereby performs a
rough (non-uniform) spectral analysis of the sound. Distributed along
Fig. 3.2 Schematic model of the auditory mechanism. (After Sachs et al. [107].)
3.2 Perception of Loudness 27
the basilar membrane are a set of inner hair cells that serve to convert
motion along the basilar membrane to neural activity. This produces
an auditory nerve representation in both time and frequency. The pro-
cessing at higher levels in the brain, shown in Figure 3.2 as a sequence
of central processing with multiple representations followed by some
type of pattern recognition, is not well understood and we can only
postulate the mechanisms used by the human brain to perceive sound
or speech. Even so, a wealth of knowledge about how sounds are per-
ceived has been discovered by careful experiments that use tones and
noise signals to stimulate the auditory system of human observers in
very specific and controlled ways. These experiments have yielded much
valuable knowledge about the sensitivity of the human auditory system
to acoustic properties such as intensity and frequency.
3.2 Perception of Loudness
A key factor in the perception of speech and other sounds is loudness.
Loudness is a perceptual quality that is related to the physical property
of sound pressure level. Loudness is quantified by relating the actual
sound pressure level of a pure tone (in dB relative to a standard refer-
ence level) to the perceived loudness of the same tone (in a unit called
phons) over the range of human hearing (20 Hz–20 kHz). This relation-
ship is shown in Figure 3.3 [37, 103]. These loudness curves show that
the perception of loudness is frequency-dependent. Specifically, the dot-
ted curve at the bottom of the figure labeled “threshold of audibility”
shows the sound pressure level that is required for a sound of a given
frequency to be just audible (by a person with normal hearing). It can
be seen that low frequencies must be significantly more intense than
frequencies in the mid-range in order that they be perceived at all. The
solid curves are equal-loudness-level contours measured by comparing
sounds at various frequencies with a pure tone of frequency 1000 Hz and
known sound pressure level. For example, the point at frequency 100 Hz
on the curve labeled 50 (phons) is obtained by adjusting the power of
the 100 Hz tone until it sounds as loud as a 1000 Hz tone having a sound
pressure level of 50 dB. Careful measurements of this kind show that a
100 Hz tone must have a sound pressure level of about 60 dB in order
28 Hearing and Auditory Perception
Fig. 3.3 Loudness level for human hearing. (After Fletcher and Munson [37].)
to be perceived to be equal in loudness to the 1000 Hz tone of sound
pressure level 50 dB. By convention, both the 50 dB 1000 Hz tone and
the 60 dB 100 Hz tone are said to have a loudness level of 50 phons
(pronounced as /F OW N Z/).
The equal-loudness-level curves show that the auditory system is
most sensitive for frequencies ranging from about 100 Hz up to about
6 kHz with the greatest sensitivity at around 3 to 4 kHz. This is almost
precisely the range of frequencies occupied by most of the sounds of
speech.
3.3 Critical Bands
The non-uniform frequency analysis performed by the basilar mem-
brane can be thought of as equivalent to that of a set of bandpass filters
whose frequency responses become increasingly broad with increas-
ing frequency. An idealized version of such a filter bank is depicted
schematically in Figure 3.4. In reality, the bandpass filters are not ideal
3.4 Pitch Perception 29
amplitude
Fig. 3.4 Schematic representation of bandpass filters according to the critical band theory
of hearing.
as shown in Figure 3.4, but their frequency responses overlap signifi-
cantly since points on the basilar membrane cannot vibrate indepen-
dently of each other. Even so, the concept of bandpass filter analysis
in the cochlea is well established, and the critical bandwidths have
been defined and measured using a variety of methods, showing that
the effective bandwidths are constant at about 100 Hz for center fre-
quencies below 500 Hz, and with a relative bandwidth of about 20%
of the center frequency above 500 Hz. An equation that fits empirical
measurements over the auditory range is
∆f
c
= 25 + 75[1 + 1.4(f
c
/1000)
2
]
0.69
, (3.1)
where ∆f
c
is the critical bandwidth associated with center frequency
f
c
[134]. Approximately 25 critical band filters span the range from
0 to 20 kHz. The concept of critical bands is very important in under-
standing such phenomena as loudness perception, pitch perception, and
masking, and it therefore provides motivation for digital representations
of the speech signal that are based on a frequency decomposition.
3.4 Pitch Perception
Most musical sounds as well as voiced speech sounds have a periodic
structure when viewed over short time intervals, and such sounds are
perceived by the auditory system as having a quality known as pitch.
Like loudness, pitch is a subjective attribute of sound that is related to
the fundamental frequency of the sound, which is a physical attribute of
the acoustic waveform [122]. The relationship between pitch (measured
30 Hearing and Auditory Perception
Fig. 3.5 Relation between subjective pitch and frequency of a pure tone.
on a nonlinear frequency scale called the mel-scale) and frequency of a
pure tone is approximated by the equation [122]:
Pitch in mels = 1127log
e
(1 + f/700), (3.2)
which is plotted in Figure 3.5. This expression is calibrated so that a
frequency of 1000 Hz corresponds to a pitch of 1000 mels. This empirical
scale describes the results of experiments where subjects were asked to
adjust the pitch of a measurement tone to half the pitch of a reference
tone. To calibrate the scale, a tone of frequency 1000 Hz is given a
pitch of 1000 mels. Below 1000 Hz, the relationship between pitch and
frequency is nearly proportional. For higher frequencies, however, the
relationship is nonlinear. For example, (3.2) shows that a frequency of
f = 5000 Hz corresponds to a pitch of 2364 mels.
The psychophysical phenomenon of pitch, as quantified by the mel-
scale, can be related to the concept of critical bands [134]. It turns out
that more or less independently of the center frequency of the band, one
critical bandwidth corresponds to about 100 mels on the pitch scale.
This is shown in Figure 3.5, where a critical band of width ∆f
c
= 160 Hz
centered on f
c
= 1000 Hz maps into a band of width 106 mels and a
critical band of width 100 Hz centered on 350 Hz maps into a band of
width 107 mels. Thus, what we know about pitch perception reinforces
the notion that the auditory system performs a frequency analysis that
can be simulated with a bank of bandpass filters whose bandwidths
increase as center frequency increases.
3.5 Auditory Masking 31
Voiced speech is quasi-periodic, but contains many frequencies. Nev-
ertheless, many of the results obtained with pure tones are relevant to
the perception of voice pitch as well. Often the term pitch period is used
for the fundamental period of the voiced speech signal even though its
usage in this way is somewhat imprecise.
3.5 Auditory Masking
The phenomenon of critical band auditory analysis can be explained
intuitively in terms of vibrations of the basilar membrane. A related
phenomenon, called masking, is also attributable to the mechanical
vibrations of the basilar membrane. Masking occurs when one sound
makes a second superimposed sound inaudible. Loud tones causing
strong vibrations at a point on the basilar membrane can swamp out
vibrations that occur nearby. Pure tones can mask other pure tones,
and noise can mask pure tones as well. A detailed discussion of masking
can be found in [134].
Figure 3.6 illustrates masking of tones by tones. The notion that a
sound becomes inaudible can be quantified with respect to the thresh-
old of audibility. As shown in Figure 3.6, an intense tone (called the
masker) tends to raise the threshold of audibility around its location on
the frequency axis as shown by the solid line. All spectral components
whose level is below this raised threshold are masked and therefore do
Masker
Shifted
threshold
Threshold
of hearing
Masked
signals
(dashed)
Unmasked
signal
log frequency
a
m
p
l
i
t
u
d
e
Fig. 3.6 Illustration of effects of masking.
32 Hearing and Auditory Perception
not need to be reproduced in a speech (or audio) processing system
because they would not be heard. Similarly, any spectral component
whose level is above the raised threshold is not masked, and therefore
will be heard. It has been shown that the masking effect is greater for
frequencies above the masking frequency than below. This is shown in
Figure 3.6 where the falloff of the shifted threshold is less abrupt above
than below the masker.
Masking is widely employed in digital representations of speech (and
audio) signals by “hiding” errors in the representation in areas where
the threshold of hearing is elevated by strong frequency components in
the signal [120]. In this way, it is possible to achieve lower data rate
representations while maintaining a high degree of perceptual fidelity.
3.6 Complete Model of Auditory Processing
In Chapter 2, we described elements of a generative model of speech
production which could, in theory, completely describe and model the
ways in which speech is produced by humans. In this chapter, we
described elements of a model of speech perception. However, the
problem that arises is that our detailed knowledge of how speech is
perceived and understood, beyond the basilar membrane processing
of the inner ear, is rudimentary at best, and thus we rely on psy-
chophysical experimentation to understand the role of loudness, critical
bands, pitch perception, and auditory masking in speech perception in
humans. Although some excellent auditory models have been proposed
[41, 48, 73, 115] and used in a range of speech processing systems, all
such models are incomplete representations of our knowledge about
how speech is understood.
4
Short-Time Analysis of Speech
In Figure 2.2 of Chapter 2, we presented a model for speech produc-
tion in which an excitation source provides the basic temporal fine
structure while a slowly varying filter provides spectral shaping (often
referred to as the spectrum envelope) to create the various sounds of
speech. In Figure 2.2, the source/system separation was presented at
an abstract level, and few details of the excitation or the linear sys-
tem were given. Both the excitation and the linear system were defined
implicitly by the assertion that the sampled speech signal was the out-
put of the overall system. Clearly, this is not sufficient to uniquely
specify either the excitation or the system. Since our goal is to extract
parameters of the model by analysis of the speech signal, it is com-
mon to assume structures (or representations) for both the excitation
generator and the linear system. One such model uses a more detailed
representation of the excitation in terms of separate source generators
for voiced and unvoiced speech as shown in Figure 4.1. In this model
the unvoiced excitation is assumed to be a random noise sequence,
and the voiced excitation is assumed to be a periodic impulse train
with impulses spaced by the pitch period (P
0
) rounded to the nearest
33
34 Short-Time Analysis of Speech
Fig. 4.1 Voiced/unvoiced/system model for a speech signal.
sample.
1
The pulses needed to model the glottal flow waveform dur-
ing voiced speech are assumed to be combined (by convolution) with
the impulse response of the linear system, which is assumed to be
slowly-time-varying (changing every 50–100 ms or so). By this we mean
that over the timescale of phonemes, the impulse response, frequency
response, and system function of the system remains relatively con-
stant. For example over time intervals of tens of milliseconds, the sys-
tem can be described by the convolution expression
s
ˆ n
[n] =

¸
m=0
h
ˆ n
[m]e
ˆ n
[n − m], (4.1)
where the subscript ˆ n denotes the time index pointing to the block of
samples of the entire speech signal s[n] wherein the impulse response
h
ˆ n
[m] applies. We use n for the time index within that interval, and m is
the index of summation in the convolution sum. In this model, the gain
G
ˆ n
is absorbed into h
ˆ n
[m] for convenience. As discussed in Chapter 2,
the most general linear time-invariant system would be characterized
by a rational system function as given in (2.1). However, to simplify
analysis, it is often assumed that the system is an all-pole system with
1
As mentioned in Chapter 3, the period of voiced speech is related to the fundamental
frequency (perceived as pitch) of the voice.
35
system function of the form:
H(z) =
G
1 −
p
¸
k=1
a
k
z
−k
. (4.2)
The coefficients G and a
k
in (4.2) change with time, and should there-
fore be indexed with the subscript ˆ n as in (4.1) and Figure 4.1, but
this complication of notation is not generally necessary since it is usu-
ally clear that the system function or difference equation only applies
over a short time interval.
2
Although the linear system is assumed to
model the composite spectrum effects of radiation, vocal tract tube,
and glottal excitation pulse shape (for voiced speech only) over a short
time interval, the linear system in the model is commonly referred
to as simply the “vocal tract” system and the corresponding impulse
response is called the “vocal tract impulse response.” For all-pole linear
systems, as represented by (4.2), the input and output are related by
a difference equation of the form:
s[n] =
p
¸
k=1
a
k
s[n − k] + Ge[n], (4.3)
where as discussed above, we have suppressed the indication of the time
at which the difference equation applies.
With such a model as the basis, common practice is to parti-
tion the analysis of the speech signal into techniques for extracting
the parameters of the excitation model, such as the pitch period and
voiced/unvoiced classification, and techniques for extracting the lin-
ear system model (which imparts the spectrum envelope or spectrum
shaping).
Because of the slowly varying nature of the speech signal, it is com-
mon to process speech in blocks (also called “frames”) over which the
properties of the speech waveform can be assumed to remain relatively
constant. This leads to the basic principle of short-time analysis, which
2
In general, we use n and m for discrete indices for sequences, but whenever we want to
indicate a specific analysis time, we use ˆ n.
36 Short-Time Analysis of Speech
is represented in a general form by the equation:
X
ˆ n
=

¸
m=−∞
T ¦x[m]w[ˆ n − m]¦, (4.4)
where X
ˆ n
represents the short-time analysis parameter (or vector of
parameters) at analysis time ˆ n.
3
The operator T¦ ¦ defines the nature
of the short-time analysis function, and w[ˆ n − m] represents a time-
shifted window sequence, whose purpose is to select a segment of the
sequence x[m] in the neighborhood of sample m = ˆ n. We will see several
examples of such operators that are designed to extract or highlight
certain features of the speech signal. The infinite limits in (4.4) imply
summation over all nonzero values of the windowed segment x
ˆ n
[m] =
x[m]w[ˆ n − m]; i.e., for all m in the region of support of the window.
For example, a finite-duration
4
window might be a Hamming window
defined by
w
H
[m] =

0.54 + 0.46cos(πm/M) −M ≤ m≤ M
0 otherwise.
(4.5)
Figure 4.2 shows a discrete-time Hamming window and its discrete-time
Fourier transform, as used throughout this chapter.
5
It can be shown
that a (2M + 1) sample Hamming window has a frequency main lobe
(full) bandwidth of 4π/M. Other windows will have similar properties,
i.e., they will be concentrated in time and frequency, and the frequency
width will be inversely proportional to the time width [89].
Figure 4.3 shows a 125 ms segment of a speech waveform that
includes both unvoiced (0–50 ms) and voiced speech (50–125 ms). Also
shown is a sequence of data windows of duration 40 ms and shifted by
15 ms (320 samples at 16 kHz sampling rate) between windows. This
illustrates how short-time analysis is implemented.
3
Sometimes (4.4) is normalized by dividing by the effective window length, i.e.,


m=−∞
w[m], so that X
ˆ n
is a weighted average.
4
An example of an infinite-duration window is w[n] = na
n
for n ≥ 0. Such a window can
lead to recursive implementations of short-time analysis functions [9].
5
We have assumed the time origin at the center of a symmetric interval of 2M + 1 samples.
A causal window would be shifted to the right by M samples.
4.1 Short-Time Energy and Zero-Crossing Rate 37
(a)
(b)
Fig. 4.2 Hamming window (a) and its discrete-time Fourier transform (b).
0 20 40 60 80 100 120
time in ms
Fig. 4.3 Section of speech waveform with short-time analysis windows.
4.1 Short-Time Energy and Zero-Crossing Rate
Two basic short-time analysis functions useful for speech signals are the
short-time energy and the short-time zero-crossing rate. These func-
tions are simple to compute, and they are useful for estimating prop-
erties of the excitation function in the model.
The short-time energy is defined as
E
ˆ n
=

¸
m=−∞
(x[m]w[ˆ n − m])
2
=

¸
m=−∞
x
2
[m]w
2
[ˆ n − m]. (4.6)
38 Short-Time Analysis of Speech
In this case the operator T¦ ¦ is simply squaring the windowed samples.
As shown in (4.6), it is often possible to express short-time analysis
operators as a convolution or linear filtering operation. In this case,
E
ˆ n
= x
2
[n] ∗ h
e
[n]

n=ˆ n
, where the impulse response of the linear filter
is h
e
[n] = w
2
[n].
Similarly, the short-time zero crossing rate is defined as the weighted
average of the number of times the speech signal changes sign within
the time window. Representing this operator in terms of linear filtering
leads to
Z
ˆ n
=

¸
m=−∞
0.5[sgn¦x[m]¦ − sgn¦x[m − 1]¦[ w[ˆ n − m], (4.7)
where
sgn¦x¦ =

1 x ≥ 0
−1 x < 0.
(4.8)
Since 0.5[sgn¦x[m]¦ − sgn¦x[m − 1]¦[ is equal to 1 if x[m] and x[m − 1]
have different algebraic signs and 0 if they have the same sign, it follows
that Z
ˆ n
in (4.7) is a weighted sum of all the instances of alternating sign
(zero-crossing) that fall within the support region of the shifted window
w[ˆ n − m]. While this is a convenient representation that fits the general
framework of (4.4), the computation of Z
ˆ n
could be implemented in
other ways.
Figure 4.4 shows an example of the short-time energy and zero-
crossing rate for a segment of speech with a transition from unvoiced
to voiced speech. In both cases, the window is a Hamming window
(two examples shown) of duration 25 ms (equivalent to 401 samples
at a 16 kHz sampling rate).
6
Thus, both the short-time energy and
the short-time zero-crossing rate are output of a lowpass filter whose
frequency response is as shown in Figure 4.2(b). For the 401-point
Hamming window used in Figure 4.4, the frequency response is very
small for discrete-time frequencies above 2π/200 rad/s (equivalent to
16000/200 = 80 Hz analog frequency). This means that the short-time
6
In the case of the short-time energy, the window applied to the signal samples was the
square-root of the Hamming window, so that he[n] = w
2
[n] is the Hamming window
defined by (4.5).
4.1 Short-Time Energy and Zero-Crossing Rate 39
0 20 40 60 80 100 120
time in ms
Fig. 4.4 Section of speech waveform with short-time energy and zero-crossing rate
superimposed.
energy and zero-crossing rate functions are slowly varying compared
to the time variations of the speech signal, and therefore, they can be
sampled at a much lower rate than that of the original speech signal.
For finite-length windows like the Hamming window, this reduction of
the sampling rate is accomplished by moving the window position ˆ n in
jumps of more than one sample as shown in Figure 4.3.
Note that during the unvoiced interval, the zero-crossing rate is rel-
atively high compared to the zero-crossing rate in the voiced interval.
Conversely, the energy is relatively low in the unvoiced region com-
pared to the energy in the voiced region. Note also that there is a small
shift of the two curves relative to events in the time waveform. This is
due to the time delay of M samples (equivalent to 12.5 ms) added to
make the analysis window filter causal.
The short-time energy and short-time zero-crossing rate are impor-
tant because they abstract valuable information about the speech sig-
nal, and they are simple to compute. The short-time energy is an indi-
cation of the amplitude of the signal in the interval around time ˆ n. From
our model, we expect unvoiced regions to have lower short-time energy
than voiced regions. Similarly, the short-time zero-crossing rate is a
crude frequency analyzer. Voiced signals have a high frequency (HF)
falloff due to the lowpass nature of the glottal pulses, while unvoiced
sounds have much more HF energy. Thus, the short-time energy and
short-time zero-crossing rate can be the basis for an algorithm for mak-
ing a decision as to whether the speech signal is voiced or unvoiced at
40 Short-Time Analysis of Speech
any particular time ˆ n. A complete algorithm would involve measure-
ments of the statistical distributions of the energy and zero-crossing
rate for both voiced and unvoiced speech segments (and also back-
ground noise distributions). These distributions can be used to derive
thresholds used in voiced/unvoiced decision [100].
4.2 Short-Time Autocorrelation Function (STACF)
The autocorrelation function is often used as a means of detecting
periodicity in signals, and it is also the basis for many spectrum analysis
methods. This makes it a useful tool for short-time speech analysis. The
STACF is defined as the deterministic autocorrelation function of the
sequence x
ˆ n
[m] = x[m]w[ˆ n − m] that is selected by the window shifted
to time ˆ n, i.e.,
φ
ˆ n
[] =

¸
m=−∞
x
ˆ n
[m]x
ˆ n
[m + ]
=

¸
m=−∞
x[m]w[ˆ n − m]x[m + ]w[ˆ n − m − ]. (4.9)
Using the familiar even-symmetric property of the autocorrelation,
φ
ˆ n
[−] = φ
ˆ n
[], (4.9) can be expressed in terms of linear time-invariant
(LTI) filtering as
φ
ˆ n
[] =

¸
m=−∞
x[m]x[m − ] ˜ w

[ˆ n − m], (4.10)
where ˜ w

[m] = w[m]w[m + ]. Note that the STACF is a two-
dimensional function of the discrete-time index ˆ n (the window position)
and the discrete-lag index . If the window has finite duration, (4.9) can
be evaluated directly or using FFT techniques (see Section 4.6). For
infinite-duration decaying exponential windows, the short-time auto-
correlation of (4.10) can be computed recursively at time ˆ n by using a
different filter ˜ w

[m] for each lag value [8, 9].
To see how the short-time autocorrelation can be used in speech
analysis, assume that a segment of the sampled speech signal is a seg-
ment of the output of the discrete-time model shown in Figure 4.1 where
4.2 Short-Time Autocorrelation Function (STACF) 41
the system is characterized at a particular analysis time by an impulse
response h[n], and the input is either a periodic impulse train or random
white noise. (Note that we have suppressed the indication of analysis
time.) Different segments of the speech signal will have the same form
of model with different excitation and system impulse response. That
is, assume that s[n] = e[n] ∗ h[n], where e[n] is the excitation to the
linear system with impulse response h[n]. A well known, and easily
proved, property of the autocorrelation function is that
φ
(s)
[] = φ
(e)
[] ∗ φ
(h)
[], (4.11)
i.e., the autocorrelation function of s[n] = e[n] ∗ h[n] is the convolution
of the autocorrelation functions of e[n] and h[n]. In the case of the
speech signal, h[n] represents the combined (by convolution) effects
of the glottal pulse shape (for voiced speech), vocal tract shape, and
radiation at the lips. For voiced speech, the autocorrelation of a peri-
odic impulse train excitation with period P
0
is a periodic impulse train
sequence with the same period. In this case, therefore, the autocorre-
lation of voiced speech is the periodic autocorrelation function
φ
(s)
[] =

¸
m=−∞
φ
(h)
[ − mP
0
]. (4.12)
In the case of unvoiced speech, the excitation can be assumed to be ran-
dom white noise, whose stochastic autocorrelation function would be
an impulse sequence at = 0. Therefore, the autocorrelation function
of unvoiced speech computed using averaging would be simply
φ
(s)
[] = φ
(h)
[]. (4.13)
Equation (4.12) assumes periodic computation of an infinite periodic
signal, and (4.13) assumes probability average or averaging over an
infinite time interval (for stationary random signals). However, the
deterministic autocorrelation function of a finite-length segment of the
speech waveform will have properties similar to those of (4.12) and
(4.13) except that the correlation values will taper off with lag due
to the tapering of the window and the fact that less and less data is
involved in the computation of the short-time autocorrelation for longer
42 Short-Time Analysis of Speech
0 10 20 30 40
−0.2
−0.1
0
0.1
0.2
0 10 20 30 40
−0.2
−0.1
0
0.1
0.2
0 10 20 30 40
−0.5
0
0.5
1
0 20 40
−0.5
0
0.5
1
(a) Voiced Segment (b) Voiced Autocorrelation
(c) Unvoiced Segment (d) Unvoiced Autocorrelation
time in ms time in ms
Fig. 4.5 Voiced and unvoiced segments of speech and their corresponding STACF.
lag values. This tapering off in level is depicted in Figure 4.5 for both a
voiced and an unvoiced speech segment. Note the peak in the autocor-
relation function for the voiced segment at the pitch period and twice
the pitch period, and note the absence of such peaks in the autocorrela-
tion function for the unvoiced segment. This suggests that the STACF
could be the basis for an algorithm for estimating/detecting the pitch
period of speech. Usually such algorithms involve the autocorrelation
function and other short-time measurements such as zero-crossings and
energy to aid in making the voiced/unvoiced decision.
Finally, observe that the STACF implicitly contains the short-time
energy since
E
ˆ n
=

¸
m=−∞
(x[m]w[ˆ n − m])
2
= φ
ˆ n
[0]. (4.14)
4.3 Short-Time Fourier Transform (STFT)
The short-time analysis functions discussed so far are examples of
the general short-time analysis principle that is the basis for most
4.3 Short-Time Fourier Transform (STFT) 43
algorithms for speech processing. We now turn our attention to what is
perhaps the most important basic concept in digital speech processing.
In subsequent chapters, we will find that the STFT defined as
X
ˆ n
(e
j ˆ ω
) =

¸
m=−∞
x[m]w[ˆ n − m]e
−j ˆ ωm
, (4.15)
is the basis for a wide range of speech analysis, coding and synthesis sys-
tems. By definition, for fixed analysis time ˆ n, the STFT is the discrete-
time Fourier transform (DTFT) of the signal x
ˆ n
[m] = x[m]w[ˆ n − m],
i.e., the DTFT of the (usually finite-duration) signal selected and
amplitude-weighted by the sliding window w[ˆ n − m] [24, 89, 129]. Thus,
the STFT is a function of two variables; ˆ n the discrete-time index denot-
ing the window position, and ˆ ω representing the analysis frequency.
7
Since (4.15) is a sequence of DTFTs, the two-dimensional function
X
ˆ n
(e
j ˆ ω
) at discrete-time ˆ n is a periodic function of continuous radian
frequency ˆ ω with period 2π [89].
As in the case of the other short-time analysis functions discussed
in this chapter, the STFT can be expressed in terms of a linear fil-
tering operation. For example, (4.15) can be expressed as the discrete
convolution
X
ˆ n
(e
j ˆ ω
) = (x[n]e
−j ˆ ωn
) ∗ w[n]

n=ˆ n
, (4.16)
or, alternatively,
X
ˆ n
(e
j ˆ ω
) =

x[n] ∗ (w[n]e
j ˆ ωn
)

e
−j ˆ ωn

n=ˆ n
. (4.17)
Recall that a typical window like a Hamming window, when viewed
as a linear filter impulse response, has a lowpass frequency response
with the cutoff frequency varying inversely with the window length.
(See Figure 4.2(b).) This means that for a fixed value of ˆ ω, X
ˆ n
(e
j ˆ ω
)
is slowly varying as ˆ n varies. Equation (4.16) can be interpreted as
follows: the amplitude modulation x[n]e
−j ˆ ωn
shifts the spectrum of
7
As before, we use ˆ n to specify the analysis time, and ˆ ω is used to distinguish the STFT
analysis frequency from the frequency variable of the non-time-dependent Fourier trans-
form frequency variable ω.
44 Short-Time Analysis of Speech
x[n] down by ˆ ω, and the window (lowpass) filter selects the resulting
band of frequencies around zero frequency. This is, of course, the
band of frequencies of x[n] that were originally centered on analysis
frequency ˆ ω. An identical conclusion follows from (4.17): x[n] is the
input to a bandpass filter with impulse response w[n]e
j ˆ ωn
, which selects
the band of frequencies centered on ˆ ω. Then that band of frequencies is
shifted down by the amplitude modulation with e
−j ˆ ωn
, resulting again
in the same lowpass signal [89].
In summary, the STFT has three interpretations: (1) It is a sequence
of discrete-time Fourier transforms of windowed signal segments, i.e., a
periodic function of ˆ ω at each window position ˆ n. (2) For each frequency
ˆ ω with ˆ n varying, it is the time sequence output of a lowpass filter that
follows frequency down-shifting by ˆ ω. (3) For each frequency ˆ ω, it is
the time sequence output resulting from frequency down-shifting the
output of a bandpass filter.
4.4 Sampling the STFT in Time and Frequency
As defined in (4.15), the STFT is a function of a continuous analysis
frequency ˆ ω. The STFT becomes a practical tool for both analysis
and applications when it is implemented with a finite-duration window
moved in steps of R > 1 samples in time and computed at a discrete
set of frequencies as in
X
rR
[k] =
rR
¸
m=rR−L+1
x[m]w[rR − m]e
−j(2πk/N)m
k = 0, 1, . . . , N − 1,
(4.18)
where N is the number of uniformly spaced frequencies across the
interval 0 ≤ ˆ ω < 2π, and L is the window length (in samples). Note
that we have assumed that w[m] is causal and nonzero only in the
range 0 ≤ m≤ L − 1 so that the windowed segment x[m]w[rR − m]
is nonzero over rR − L + 1 ≤ m≤ rR. To aid in interpretation, it is
helpful to write (4.18) in the equivalent form:
X
rR
[k] =
˜
X
rR
[k]e
−j(2πk/N)rR
k = 0, 1, . . . , N − 1, (4.19)
4.4 Sampling the STFT in Time and Frequency 45
where
˜
X
rR
[k] =
L−1
¸
m=0
x[rR − m]w[m]e
j(2πk/N)m
k = 0, 1, . . . , N − 1. (4.20)
Since we have assumed, for specificity, that w[m] = 0 only in the range
0 ≤ m≤ L − 1, the alternative form,
˜
X
rR
[k], has the interpretation of
an N-point DFT of the sequence x[rR − m]w[m], which, due to the
definition of the window, is nonzero in the interval 0 ≤ m≤ L − 1.
8
In
(4.20), the analysis time rR is shifted to the time origin of the DFT
computation, and the segment of the speech signal is the time-reversed
sequence of L samples that precedes the analysis time. The complex
exponential factor e
−j(2πk/N)rR
in (4.19) results from the shift of the
time origin.
For a discrete-time Fourier transform interpretation, it follows that
˜
X
rR
[k] can be computed by the following process:
(1) Form the sequence x
rR
[m] = x[rR − m]w[m], for m =
0, 1, . . . , L − 1.
(2) Compute the complex conjugate of the N-point DFT of
the sequence x
rR
[m]. (This can be done efficiently with an
N-point FFT algorithm.)
(3) The multiplication by e
−j(2πk/N)rR
can be done if necessary,
but often can be omitted (as in computing a sound spectro-
gram or spectrographic display).
(4) Move the time origin by R samples (i.e., r →r + 1) and
repeat steps (1), (2), and (3), etc.
The remaining issue for complete specification of the sampled STFT
is specification of the temporal sampling period, R, and the number of
uniform frequencies, N. It can easily be shown that both R and N
are determined entirely by the time width and frequency bandwidth of
the lowpass window, w[m], used to compute the STFT [1], giving the
8
The DFT is normally defined with a negative exponent. Thus, since x[rR − m]w[m] is
real, (4.20) is the complex conjugate of the DFT of the windowed sequence [89].
46 Short-Time Analysis of Speech
following constraints on R and N:
(1) R ≤ L/(2C) where C is a constant that is dependent on the
window frequency bandwidth; C = 2 for a Hamming window,
C = 1 for a rectangular window.
(2) N ≥ L, where L is the window length in samples.
Constraint (1) above is related to sampling the STFT in time at a
rate of twice the window bandwidth in frequency in order to eliminate
aliasing in frequency of the STFT, and constraint (2) above is related
to sampling in frequency at a rate of twice the equivalent time width
of the window to ensure that there is no aliasing in time of the STFT.
4.5 The Speech Spectrogram
Since the 1940s, the sound spectrogram has been a basic tool for gain-
ing understanding of how the sounds of speech are produced and how
phonetic information is encoded in the speech signal. Up until the
1970s, spectrograms were made by an ingenious device comprised of
an audio tape loop, variable analog bandpass filter, and electrically
sensitive paper [66]. Today spectrograms like those in Figure 4.6 are
made by DSP techniques [87] and displayed as either pseudo-color or
gray-scale images on computer screens.
Sound spectrograms like those in Figure 4.6 are simply a display of
the magnitude of the STFT. Specifically, the images in Figure 4.6 are
plots of
S(t
r
, f
k
) = 20log
10
[
˜
X
rR
[k][ = 20log
10
[X
rR
[k][, (4.21)
where the plot axes are labeled in terms of analog time and fre-
quency through the relations t
r
= rRT and f
k
= k/(NT), where T is
the sampling period of the discrete-time signal x[n] = x
a
(nT). In order
to make smooth looking plots like those in Figure 4.6, R is usually
quite small compared to both the window length L and the num-
ber of samples in the frequency dimension, N, which may be much
larger than the window length L. Such a function of two variables
can be plotted on a two dimensional surface (such as this text) as
either a gray-scale or a color-mapped image. Figure 4.6 shows the time
4.5 The Speech Spectrogram 47
−1
0
1
SH UH D W IY CH EY S
f
r
e
q
u
e
n
c
y

i
n

k
H
z
0 100 200 300 400 500 600
0
1
2
3
4
5
−100
−50
0
f
r
e
q
u
e
n
c
y

i
n

k
H
z
time in ms
0 100 200 300 400 500 600
0
1
2
3
4
5
−120
−100
−80
−60
−40
−20
0
Fig. 4.6 Spectrogram for speech signal of Figure 1.1.
waveform at the top and two spectrograms computed with different
length analysis windows. The bars on the right calibrate the color
map (in dB).
A careful interpretation of (4.20) and the corresponding spectro-
gram images leads to valuable insight into the nature of the speech
signal. First note that the window sequence w[m] is nonzero only over
the interval 0 ≤ m≤ L − 1. The length of the window has a major
effect on the spectrogram image. The upper spectrogram in Figure 4.6
was computed with a window length of 101 samples, corresponding to
10 ms time duration. This window length is on the order of the length
of a pitch period of the waveform during voiced intervals. As a result,
in voiced intervals, the spectrogram displays vertically oriented stri-
ations corresponding to the fact that the sliding window sometimes
includes mostly large amplitude samples, then mostly small amplitude
48 Short-Time Analysis of Speech
samples, etc. As a result of the short analysis window, each individ-
ual pitch period is resolved in the time dimension, but the resolution
in the frequency dimension is poor. For this reason, if the analysis
window is short, the spectrogram is called a wide-band spectrogram.
This is consistent with the linear filtering interpretation of the STFT,
since a short analysis filter has a wide passband. Conversely, when the
window length is long, the spectrogram is a narrow-band spectrogram,
which is characterized by good frequency resolution and poor time
resolution.
The upper plot in Figure 4.7, for example, shows S(t
r
, f
k
) as a
function of f
k
at time t
r
= 430 ms. This vertical slice through the
spectrogram is at the position of the black vertical line in the upper
spectrogram of Figure 4.6. Note the three broad peaks in the spec-
trum slice at time t
r
= 430 ms, and observe that similar slices would
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
−80
−60
−40
−20
0
l
o
g

m
a
g
n
i
t
u
d
e

(
d
B
)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
−80
−60
−40
−20
0
frequency in kHz
l
o
g

m
a
g
n
i
t
u
d
e

(
d
B
)
Fig. 4.7 Short-time spectrum at time 430 ms (dark vertical line in Figure 4.6) with Hamming
window of length M = 101 in upper plot and M = 401 in lower plot.
4.6 Relation of STFT to STACF 49
be obtained at other times around t
r
= 430 ms. These large peaks are
representative of the underlying resonances of the vocal tract at the
corresponding time in the production of the speech signal.
The lower spectrogram in Figure 4.6 was computed with a win-
dow length of 401 samples, corresponding to 40 ms time duration. This
window length is on the order of several pitch periods of the waveform
during voiced intervals. As a result, the spectrogram no longer displays
vertically oriented striations since several periods are included in the
window no matter where it is placed on the waveform in the vicinity
of the analysis time t
r
. As a result, the spectrogram is not as sensitive
to rapid time variations, but the resolution in the frequency dimension
is much better. Therefore, the striations tend to be horizontally ori-
ented in the narrow-band case since the fundamental frequency and its
harmonics are all resolved. The lower plot in Figure 4.7, for example,
shows S(t
r
, f
k
) as a function of f
k
at time t
r
= 430 ms. In this case,
the signal is voiced at the position of the window, so over the analy-
sis window interval it acts very much like a periodic signal. Periodic
signals have Fourier spectra that are composed of impulses at the fun-
damental frequency, f
0
, and at integer multiples (harmonics) of the
fundamental frequency [89]. Multiplying by the analysis window in the
time-domain results in convolution of the Fourier transform of the win-
dow with the impulses in the spectrum of the periodic signal [89]. This
is evident in the lower plot of Figure 4.7, where the local maxima of the
curve are spaced at multiples of the fundamental frequency f
0
= 1/T
0
,
where T
0
is the fundamental period (pitch period) of the signal.
9
4.6 Relation of STFT to STACF
A basic property of Fourier transforms is that the inverse Fourier trans-
form of the magnitude-squared of the Fourier transform of a signal is the
autocorrelation function for that signal [89]. Since the STFT defined in
(4.15) is a discrete-time Fourier transform for fixed window position,
it follows that φ
ˆ n
[] given by (4.9) is related to the STFT given by
9
Each “ripple” in the lower plot is essentially a frequency-shifted copy of the Fourier trans-
form of the Hamming window used in the analysis.
50 Short-Time Analysis of Speech
(4.15) as
φ
ˆ n
[] =
1

π
−π
[X
ˆ n
(e
j ˆ ω
)[
2
e
j ˆ ω
dˆ ω. (4.22)
The STACF can also be computed from the sampled STFT. In partic-
ular, an inverse DFT can be used to compute
˜
φ
rR
[] =
1
N
N−1
¸
k=0
[
˜
X
rR
(e
j(2πk/N)
)[
2
e
j(2πk/N)
(4.23)
and
˜
φ
rR
[] = φ
rR
[] for = 0, 1, . . . , L − 1 if N ≥ 2L. If L < N < 2L
time aliasing will occur, but
˜
φ
rR
[] = φ
rR
[] for = 0, 1, . . . , N − L. Note
that (4.14) shows that the short-time energy can also be obtained from
either (4.22) or (4.23) by setting = 0.
Figure 4.8 illustrates the equivalence between the STACF and the
STFT. Figures 4.8(a) and 4.8(c) show the voiced and unvoiced autocor-
relation functions that were shown in Figures 4.5(b) and 4.5(d). On the
right are the corresponding STFTs. The peaks around 9 and 18 ms in
0 10 20 30 40
−0.5
0
0.5
1
0 10 20 30 40
−0.5
0
0.5
1
0 2000 4000 6000 8000
−100
−50
0
50
0 2000 4000 6000 8000
−100
−50
0
50
(b) Voiced STFT (a) Voiced Autocorrelation
(d) Unvoiced STFT (c) Unvoiced Autocorrelation
time in msec frequency in kHz
Fig. 4.8 STACF and corresponding STFT.
4.7 Short-Time Fourier Synthesis 51
the autocorrelation function of the voiced segment imply a fundamen-
tal frequency at this time of approximately 1000/9 = 111 Hz. Similarly,
note in the STFT on the right in Figure 4.8(b) that there are approx-
imately 18 local regularly spaced peaks in the range 0–2000 Hz. Thus,
we can estimate the fundamental frequency to be 2000/18 = 111 Hz as
before.
4.7 Short-Time Fourier Synthesis
From the linear filtering point of view of the STFT, (4.19) and (4.20)
represent the downsampled (by the factor R) output of the process
of bandpass filtering with impulse response w[n]e
j(2πk/N)n
followed
by frequency-down-shifting by (2πk/N). Alternatively, (4.18) shows
that X
rR
[k] is a downsampled output of the lowpass window filter
with frequency-down-shifted input x[n]e
−j(2πk/N)n
. The left half of Fig-
ure 4.9 shows a block diagram representation of the STFT as a com-
bination of modulation, followed by lowpass filtering, followed by a
downsampler by R. This structure is often called a filter bank, and
the outputs of the individual filters are called the channel signals. The
w[n]
w[n]
w[n]
x
x
x x
x
x
S
h
o
r
t

T
i
m
e

M
o
d
i
f
i
c
a
t
i
o
n
s
R
R
R
R
R
R
f [n]
f [n]
f [n]
+
] [k X
n
] [n x
n N j
N
e
) 1 (
2
− −
π
n N j
N
e
) 1 (
2

π
n j
N
e
) 1 (


n j
N
e
) 0 (


n j
N
e
) 1 (

n j
N
e
) 0 (

] [n y
] [k X
rR
] [k Y
rR
Fig. 4.9 Filterbank interpretation of short-time Fourier analysis and synthesis.
52 Short-Time Analysis of Speech
spectrogram is comprised of the set of outputs of the filter bank with
each channel signal corresponding to a horizontal line in the spectro-
gram. If we want to use the STFT for other types of speech process-
ing, we may need to consider if and how it is possible to reconstruct
the speech signal from the STFT, i.e., from the channel signals. The
remainder of Figure 4.9 shows how this can be done.
First, the diagram shows the possibility that the STFT might be
modified by some sort of processing. An example might be quantization
of the channel signals for data compression. The modified STFT is
denoted as Y
rR
[k]. The remaining parts of the structure on the right in
Figure 4.9 implement the synthesis of a new time sequence y[n] from
the STFT. This part of the diagram represents the synthesis equation
y[n] =
N−1
¸
k=0


¸
r=−∞
Y
rR
[k]f[n − rR]

e
j(2πk/N)n
. (4.24)
The steps represented by (4.24) and the right half of Figure 4.9 involve
first upsampling by R followed by linear filtering with f[n] (called
the synthesis window). This operation, defined by the part within the
parenthesis in (4.24), interpolates the STFT Y
rR
[k] to the time sam-
pling rate of the speech signal for each channel k.
10
Then, synthesis
is achieved by modulation with e
j(2πk/N)n
, which up-shifts the lowpass
interpolated channel signals back to their original frequency bands cen-
tered on frequencies (2πk/N). The sum of the up-shifted signals is the
synthesized output.
There are many variations on the short-time Fourier analy-
sis/synthesis paradigm. The FFT algorithm can be used to implement
both analysis and synthesis; special efficiencies result when L > R = N,
and the analysis and synthesis can be accomplished using only real fil-
tering and with modulation operations being implicitly achieved by
the downsampling [24, 129]. However, it is most important to note
that with careful choice of the parameters L, R, and N and careful
design of the analysis window w[m] together with the synthesis win-
dow f[n], it is possible to reconstruct the signal x[n] with negligible
10
The sequence f[n] is the impulse response of a LTI filter with gain R/N and normalized
cutoff frequency π/R. It is often a scaled version of w[n].
4.8 Short-Time Analysis is Fundamental to our Thinking 53
error (y[n] = x[n]) from the unmodified STFT Y
rR
[k] = X
rR
[k]. One
condition that guarantees exact reconstruction is [24, 129].

¸
r=−∞
w[rR − n + qN]f[n − rR] =

1 q = 0
0 q = 0.
(4.25)
Short-time Fourier analysis and synthesis is generally formulated
with equally spaced channels with equal bandwidths. However, we
have seen that models for auditory processing involve nonuniform filter
banks. Such filter banks can be implemented as tree structures where
frequency bands are successively divided into low- and HF bands. Such
structures are essentially the same as wavelet decompositions, a topic
beyond the scope of this text, but one for which a large body of liter-
ature exists. See, for example [18, 125, 129].
4.8 Short-Time Analysis is Fundamental to our Thinking
It can be argued that the short-time analysis principle, and particu-
larly the short-time Fourier representation of speech, is fundamental to
our thinking about the speech signal and it leads us to a wide variety
of techniques for achieving our goal of moving from the sampled time
waveform back along the speech chain toward the implicit message.
The fact that almost perfect reconstruction can be achieved from the
filter bank channel signals gives the short-time Fourier representation
major credibility in the digital speech processing tool kit. This impor-
tance is strengthened by the fact that, as discussed briefly in Chapter 3,
models for auditory processing are based on a filter bank as the first
stage of processing. Much of our knowledge of perceptual effects is
framed in terms of frequency analysis, and thus, the STFT representa-
tion provides a natural framework within which this knowledge can be
represented and exploited to obtain efficient representations of speech
and more general audio signals. We will have more to say about this
in Chapter 7, but first we will consider (in Chapter 5) the technique
of cepstrum analysis, which is based directly on the STFT, and the
technique of linear predictive analysis (in Chapter 6), which is based
on the STACF and thus equivalently on the STFT.
5
Homomorphic Speech Analysis
The STFT provides a useful framework for thinking about almost all
the important techniques for analysis of speech that have been devel-
oped so far. An important concept that flows directly from the STFT
is the cepstrum, more specifically, the short-time cepstrum, of speech.
In this chapter, we explore the use of the short-time cepstrum as a
representation of speech and as a basis for estimating the parameters
of the speech generation model. A more detailed discussion of the uses
of the cepstrum in speech processing can be found in [110].
5.1 Definition of the Cepstrum and Complex Cepstrum
The cepstrum was defined by Bogert, Healy, and Tukey to be the inverse
Fourier transform of the log magnitude spectrum of a signal [16]. Their
original definition, loosely framed in terms of spectrum analysis of ana-
log signals, was motivated by the fact that the logarithm of the Fourier
spectrum of a signal containing an echo has an additive periodic com-
ponent depending only on the echo size and delay, and that further
Fourier analysis of the log spectrum can aid in detecting the presence
54
5.1 Definition of the Cepstrum and Complex Cepstrum 55
of that echo. Oppenheim, Schafer, and Stockham showed that the cep-
strum is related to the more general concept of homomorphic filtering
of signals that are combined by convolution [85, 90, 109]. They gave a
definition of the cepstrum of a discrete-time signal as
c[n] =
1

π
−π
log[X(e

)[e
jωn
dω, (5.1a)
where log[X(e

)[ is the logarithm of the magnitude of the DTFT
of the signal, and they extended the concept by defining the complex
cepstrum as
ˆ x[n] =
1

π
−π
log¦X(e

)¦e
jωn
dω, (5.1b)
where log¦X(e

)¦ is the complex logarithm of X(e

) defined as
ˆ
X(e

) = log¦X(e

)¦ = log[X(e

)[ + j arg[X(e

)]. (5.1c)
The transformation implied by (5.1b) is depicted as the block diagram
in Figure 5.1. The same diagram represents the cepstrum if the complex
logarithm is replaced by the logarithm of the magnitude of the DTFT.
Since we restrict our attention to real sequences, x[n], it follows from
the symmetry properties of Fourier transforms that the cepstrum is the
even part of the complex cepstrum, i.e., c[n] = (ˆ x[n] + ˆ x[−n])/2 [89].
As shown in Figure 5.1, the operation of computing the complex cep-
strum from the input can be denoted as ˆ x[n] = D

¦x[n]¦. In the theory
of homomorphic systems D

¦ ¦ is called the characteristic system for
convolution. The connection between the cepstrum concept and homo-
morphic filtering of convolved signals is that the complex cepstrum has
the property that if x[n] = x
1
[n] ∗ x
2
[n], then
ˆ x[n] = D

¦x
1
[n] ∗ x
2
[n]¦ = ˆ x
1
[n] + ˆ x
2
[n]. (5.2)
Fig. 5.1 Computing the complex cepstrum using the DTFT.
56 Homomorphic Speech Analysis
Fig. 5.2 The inverse of the characteristic system for convolution (inverse complex cepstrum).
That is, the complex cepstrum operator transforms convolution into
addition. This property, which is true for both the cepstrum and the
complex cepstrum, is what makes the cepstrum and the complex cep-
strum useful for speech analysis, since our model for speech produc-
tion involves convolution of the excitation with the vocal tract impulse
response, and our goal is often to separate the excitation signal from
the vocal tract signal. In the case of the complex cepstrum, the inverse
of the characteristic system exists as in Figure 5.2, which shows the
reverse cascade of the inverses of the operators in Figure 5.1. Homo-
morphic filtering of convolved signals is achieved by forming a modified
complex cepstrum
ˆ y[n] = g[n]ˆ x[n], (5.3)
where g[n] is a window (a “lifter” in the terminology of Bogert et al.)
which selects a portion of the complex cepstrum for inverse processing.
A modified output signal y[n] can then be obtained as the output of
Figure 5.2 with ˆ y[n] given by (5.3) as input. Observe that (5.3) defines
a linear operator in the conventional sense, i.e., if ˆ x[n] = ˆ x
1
[n] + ˆ x
2
[n]
then ˆ y[n] = g[n]ˆ x
1
[n] + g[n]ˆ x
2
[n]. Therefore, the output of the inverse
characteristic system will have the form y[n] = y
1
[n] ∗ y
2
[n], where
ˆ y
1
[n] = g[n]ˆ x
1
[n] is the complex cepstrum of y
1
[n], etc. Examples of
lifters used in homomorphic filtering of speech are given in Sections 5.4
and 5.6.2.
The key issue in the definition and computation of the complex cep-
strum is the computation of the complex logarithm; more specifically,
the computation of the phase angle arg[X(e

)], which must be done
so as to preserve an additive combination of phases for two signals
combined by convolution [90, 109].
5.2 The Short-Time Cepstrum 57
The independent variable of the cepstrum and complex cepstrum
is nominally time. The crucial observation leading to the cepstrum
concept is that the log spectrum can be treated as a waveform to be
subjected to further Fourier analysis. To emphasize this interchanging
of domains of reference, Bogert et al. [16] coined the word cepstrum
by transposing some of the letters in the word spectrum. They created
many other special terms in this way including quefrency as the name
for the independent variable of the cepstrum and liftering for the oper-
ation of linear filtering the log magnitude spectrum by the operation
of (5.3). Only the terms cepstrum, quefrency, and liftering are widely
used today.
5.2 The Short-Time Cepstrum
The application of these definitions to speech requires that the
DTFT be replaced by the STFT. Thus the short-time cepstrum is
defined as
c
ˆ n
[m] =
1

π
−π
log[X
ˆ n
(e
j ˆ ω
)[e
j ˆ ωm
dˆ ω, (5.4)
where X
ˆ n
(e
j ˆ ω
) is the STFT defined in (4.15), and the short-time
complex cepstrum is likewise defined by replacing X(e
j ˆ ω
) by X
ˆ n
(e
j ˆ ω
)
in (5.1b).
1
The similarity to the STACF defined by (4.9) should be
clear. The short-time cepstrum is a sequence of cepstra of windowed
finite-duration segments of the speech waveform. By analogy, a “cep-
strogram” would be an image obtained by plotting the magnitude of
the short-time cepstrum as a function of quefrency m and analysis
time ˆ n.
5.3 Computation of the Cepstrum
In (5.1a) and (5.1b), the cepstrum and complex cepstrum are defined
in terms of the DTFT. This is useful in the basic definitions, but not
1
In cases where we wish to explicitly indicate analysis-time dependence of the short-time
cepstrum, we will use ˆ n for the analysis time and m for quefrency as in (5.4), but as in
other instances, we often suppress the subscript ˆ n.
58 Homomorphic Speech Analysis
for use in processing sampled speech signals. Fortunately, several com-
putational options exist.
5.3.1 Computation Using the DFT
Since the DFT (computed with an FFT algorithm) is a sampled
(in frequency) version of the DTFT of a finite-length sequence (i.e.,
X[k] = X(e
j2πk/N
) [89]), the DFT and inverse DFT can be substi-
tuted for the DTFT and its inverse in (5.1a) and (5.1b) as shown in
Figure 5.3, which shows that the complex cepstrum can be computed
approximately using the equations
X[k] =
N−1
¸
n=0
x[n]e
−j(2πk/N)n
(5.5a)
ˆ
X[k] = log[X[k][ + j arg¦X[k]¦ (5.5b)
˜
ˆ x[n] =
1
N
N−1
¸
k=0
ˆ
X[k]e
j(2πk/N)n
. (5.5c)
Note the “tilde” ˜ symbol above ˆ x[n] in (5.5c) and in Figure 5.3. Its
purpose is to emphasize that using the DFT instead of the DTFT
results in an approximation to (5.1b) due to the time-domain aliasing
resulting from the sampling of the log of the DTFT [89]. That is,
˜
ˆ x[n] =

¸
r=−∞
ˆ x[n + rN], (5.6)
where ˆ x[n] is the complex cepstrum defined by (5.1b). An identical
equation holds for the time-aliased cepstrum ˜ c[n].
The effect of time-domain aliasing can be made negligible by using a
large value for N. A more serious problem in computation of the com-
plex cepstrum is the computation of the complex logarithm. This is
Fig. 5.3 Computing the cepstrum or complex cepstrum using the DFT.
5.3 Computation of the Cepstrum 59
because the angle of a complex number is usually specified modulo
2π, i.e., by the principal value. In order for the complex cepstrum
to be evaluated properly, the phase of the sampled DTFT must be
evaluated as samples of a continuous function of frequency. If the princi-
pal value phase is first computed at discrete frequencies, then it must be
“unwrapped” modulo 2π in order to ensure that convolutions are trans-
formed into additions. A variety of algorithms have been developed for
phase unwrapping [90, 109, 127]. While accurate phase unwrapping
presents a challenge in computing the complex cepstrum, it is not a
problem in computing the cepstrum, since the phase is not used. Fur-
thermore, phase unwrapping can be avoided by using numerical com-
putation of the poles and zeros of the z-transform.
5.3.2 z-Transform Analysis
The characteristic system for convolution can also be represented by
the two-sided z-transform as depicted in Figure 5.4. This is very useful
for theoretical investigations, and recent developments in polynomial
root finding [117] have made the z-transform representation a viable
computational basis as well. For this purpose, we assume that the input
signal x[n] has a rational z-transform of the form:
X(z) = X
max
(z)X
uc
(z)X
min
(z), (5.7)
where
X
max
(z) = z
Mo
Mo
¸
k=1
(1 − a
k
z
−1
) =
Mo
¸
k=1
(−a
k
)
Mo
¸
k=1
(1 − a
−1
k
z), (5.8a)
X
uc
(z) =
Muc
¸
k=1
(1 − e

k
z
−1
), (5.8b)
X
min
(z) = A
M
i
¸
k=1
(1 − b
k
z
−1
)
N
i
¸
k=1
(1 − c
k
z
−1
)
. (5.8c)
The zeros of X
max
(z), i.e., z
k
= a
k
, are zeros of X(z) outside of the
unit circle ([a
k
[ > 1). X
max
(z) is thus the maximum-phase part of X(z).
60 Homomorphic Speech Analysis
Fig. 5.4 z-transform representation of characteristic system for convolution.
X
uc
(z) contains all the zeros (with angles θ
k
) on the unit circle. The
minimum-phase part is X
min
(z), where b
k
and c
k
are zeros and poles,
respectively, that are inside the unit circle ([b
k
[ < 1 and [c
k
[ < 1). The
factor z
Mo
implies a shift of M
o
samples to the left. It is included to
simplify the results in (5.11).
The complex cepstrum of x[n] is determined by assuming that the
complex logarithm log¦X(z)¦ results in the sum of logarithms of each
of the product terms, i.e.,
ˆ
X(z) = log

Mo
¸
k=1
(−a
k
)

+
Mo
¸
k=1
log(1 − a
−1
k
z) +
Muc
¸
k=1
log(1 − e

k
z
−1
)
+ log[A[ +
M
i
¸
k=1
log(1 − b
k
z
−1
) −
N
i
¸
k=1
log(1 − c
k
z
−1
). (5.9)
Applying the power series expansion
log(1 − a) = −

¸
n=1
a
n
n
[a[ < 1 (5.10)
to each of the terms in (5.9) and collecting the coefficients of the positive
and negative powers of z gives
ˆ x[n] =

Mo
¸
k=1
a
n
k
n
n < 0
log[A[ + log

mo
¸
k=1
(−a
k
)

n = 0


Muc
¸
k=1
e

k
n
n

M
i
¸
k=1
b
n
k
n
+
N
i
¸
k=1
c
n
k
n

n > 0.
(5.11)
Given all the poles and zeros of a z-transform X(z), (5.11) allows
us to compute the complex cepstrum with no approximation. This is
5.3 Computation of the Cepstrum 61
the case in theoretical analysis where the poles and zeros are specified.
However, (5.11) is also useful as the basis for computation. All that is
needed is a process for obtaining the z-transform as a rational function
and a process for finding the zeros of the numerator and denominator.
This has become more feasible with increasing computational power
and with new advances in finding roots of large polynomials [117].
One method of obtaining a z-transform is simply to select a finite-
length sequence of samples of a signal. The z-transform is then simply
a polynomial with the samples x[n] as coefficients, i.e.,
X(z) =
M
¸
n=0
x[n]z
−n
= A
Mo
¸
k=1
(1 − a
k
z
−1
)
M
i
¸
k=1
(1 − b
k
z
−1
). (5.12)
A second method that yields a z-transform is the method of linear
predictive analysis, to be discussed in Chapter 6.
5.3.3 Recursive Computation of the Complex Cepstrum
Another approach to computing the complex cepstrum applies only to
minimum-phase signals, i.e., signals having a z-transform whose poles
and zeros are inside the unit circle. An example would be the impulse
response of an all-pole vocal tract model with system function
H(z) =
G
1 −
p
¸
k=1
α
k
z
−k
=
G
p
¸
k=1
(1 − c
k
z
−1
)
. (5.13)
Such models are implicit in the use of linear predictive analysis of speech
(Chapter 6). In this case, all the poles c
k
must be inside the unit circle
for stability of the system. From (5.11) it follows that the complex
cepstrum of the impulse response h[n] corresponding to H(z) is
ˆ
h[n] =

0 n < 0
log[G[ n = 0
p
¸
k=1
c
n
k
n
n > 0.
(5.14)
62 Homomorphic Speech Analysis
It can be shown [90] that the impulse response and its complex cep-
strum are related by the recursion formula:
ˆ
h[n] =

0 n < 0
logG n = 0
h[n]
h[0]

n−1
¸
k=1

k
n

ˆ
h[k]h[n − k]
h[0]
n ≥ 1.
(5.15)
Furthermore, working with the reciprocal (negative logarithm) of
(5.13), it can be shown that there is a direct recursive relationship
between the coefficients of the denominator polynomial in (5.13) and
the complex cepstrum of the impulse response of the model filter, i.e.,
ˆ
h[n] =

0 n < 0
logG n = 0
α
n
+
n−1
¸
k=1

k
n

ˆ
h[k]α
n−k
n > 0.
(5.16)
From (5.16) it follows that the coefficients of the denominator polyno-
mial can be obtained from the complex cepstrum through
α
n
=
ˆ
h[n] −
n−1
¸
k=1

k
n

ˆ
h[k]α
n−k
1 ≤ n ≤ p. (5.17)
From (5.17), it follows that p + 1 values of the complex cepstrum are
sufficient to fully determine the speech model system in (5.13) since all
the denominator coefficients and G can be computed from
ˆ
h[n] for n =
0, 1, . . . , p using (5.17). This fact is the basis for the use of the cepstrum
in speech coding and speech recognition as a vector representation of
the vocal tract properties of a frame of speech.
5.4 Short-Time Homomorphic Filtering of Speech
Figure 5.5 shows an example of the short-time cepstrum of speech for
the segments of voiced and unvoiced speech in Figures 4.5(a) and 4.5(c).
The low quefrency part of the cepstrum is expected to be representative
of the slow variations (with frequency) in the log spectrum, while the
5.4 Short-Time Homomorphic Filtering of Speech 63
0 10 20 30 40
−1
0
1
0 10 20 30 40
−1
0
1
0 2000 4000 6000 8000
−8
−6
−4
−2
0
2
4
0 2000 4000 6000 8000
−8
−6
−4
−2
0
2
4
(b) Voiced log STFT (a) Voiced Cepstrum
(d) Unvoiced log STFT (c) Unvoiced Cepstrum
time in ms frequency in kHz
Fig. 5.5 Short-time cepstra and corresponding STFTs and homomorphically-smoothed
spectra.
high quefrency components would correspond to the more rapid fluctu-
ations of the log spectrum. This is illustrated by the plots in Figure 5.5.
The log magnitudes of the corresponding short-time spectra are shown
on the right as Figures 5.5(b) and 5.5(d). Note that the spectrum for
the voiced segment in Figure 5.5(b) has a structure of periodic ripples
due to the harmonic structure of the quasi-periodic segment of voiced
speech. This periodic structure in the log spectrum of Figure 5.5(b),
manifests itself in the cepstrum peak at a quefrency of about 9 ms
in Figure 5.5(a). The existence of this peak in the quefrency range of
expected pitch periods strongly signals voiced speech. Furthermore, the
quefrency of the peak is an accurate estimate of the pitch period dur-
ing the corresponding speech interval. As shown in Figure 4.5(b), the
autocorrelation function also displays an indication of periodicity, but
not nearly as unambiguously as does the cepstrum. On the other hand,
note that the rapid variations of the unvoiced spectra appear random
with no periodic structure. This is typical of Fourier transforms (peri-
odograms) of short segments of random signals. As a result, there is no
strong peak indicating periodicity as in the voiced case.
64 Homomorphic Speech Analysis
To illustrate the effect of liftering, the quefrencies above 5 ms are
multiplied by zero and the quefrencies below 5 ms are multiplied by 1
(with a short transition taper as shown in Figures 5.5(a) and 5.5(c)).
The DFT of the resulting modified cepstrum is plotted as the smooth
curve that is superimposed on the short-time spectra in Figures 5.5(b)
and 5.5(d), respectively. These slowly varying log spectra clearly retain
the general spectral shape with peaks corresponding to the formant
resonance structure for the segment of speech under analysis. Therefore,
a useful perspective is that by liftering the cepstrum, it is possible
to separate information about the vocal tract contributions from the
short-time speech spectrum [88].
5.5 Application to Pitch Detection
The cepstrum was first applied in speech processing to determine the
excitation parameters for the discrete-time speech model of Figure 7.10.
Noll [83] applied the short-time cepstrum to detect local periodicity
(voiced speech) or the lack thereof (unvoiced speech). This is illus-
trated in Figure 5.6, which shows a plot that is very similar to the plot
first published by Noll [83]. On the left is a sequence of log short-time
spectra (rapidly varying curves) and on the right is the correspond-
ing sequence of cepstra computed from the log spectra on the left.
The successive spectra and cepstra are for 50 ms segments obtained by
moving the window in steps of 12.5 ms (100 samples at a sampling rate
of 8000 samples/sec). From the discussion of Section 5.4, it is apparent
that for the positions 1 through 5, the window includes only unvoiced
speech, while for positions 6 and 7 the signal within the window is
partly voiced and partly unvoiced. For positions 8 through 15 the win-
dow only includes voiced speech. Note again that the rapid variations
of the unvoiced spectra appear random with no periodic structure. On
the other hand, the spectra for voiced segments have a structure of
periodic ripples due to the harmonic structure of the quasi-periodic
segment of voiced speech. As can be seen from the plots on the right,
the cepstrum peak at a quefrency of about 11–12 ms strongly signals
voiced speech, and the quefrency of the peak is an accurate estimate of
the pitch period during the corresponding speech interval.
5.5 Application to Pitch Detection 65
0 1 2 3 4
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
0 5 10 15 20 25
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Short-Time Log Spectra in Cepstrum Analysis
frequency in kHz
Short-Time Cepstra
time in ms
w
i
n
d
o
w
n
u
m
b
e
r
Fig. 5.6 Short-time cepstra and corresponding STFTs and homomorphically-smoothed
spectra.
The essence of the pitch detection algorithm proposed by Noll is
to compute a sequence of short-time cepstra and search each succes-
sive cepstrum for a peak in the quefrency region of the expected pitch
period. Presence of a strong peak implies voiced speech, and the que-
frency location of the peak gives the estimate of the pitch period. As
in most model-based signal processing applications of concepts such
as the cepstrum, the pitch detection algorithm includes many fea-
tures designed to handle cases that do not fit the underlying model
very well. For example, for frames 6 and 7 the cepstrum peak is weak
66 Homomorphic Speech Analysis
corresponding to the transition from unvoiced to voiced speech. In other
problematic cases, the peak at twice the pitch period may be stronger
than the peak at the quefrency of the pitch period. Noll applied tem-
poral continuity constraints to prevent such errors.
5.6 Applications to Pattern Recognition
Perhaps the most pervasive application of the cepstrum in speech pro-
cessing is its use in pattern recognition systems such as design of vec-
tor quantizers (VQ) and automatic speech recognizers (ASR). In such
applications, a speech signal is represented on a frame-by-frame basis
by a sequence of short-time cepstra. As we have shown, cepstra can be
computed either by z-transform analysis or by DFT implementation
of the characteristic system. In either case, we can assume that the
cepstrum vector corresponds to a gain-normalized (c[0] = 0) minimum-
phase vocal tract impulse response that is defined by the complex
cepstrum
2
ˆ
h[n] =

2c[n] 1 ≤ n ≤ n
co
0 n < 0.
(5.18)
In problems such as VQ or ASR, a test pattern c[n] (vector of cep-
strum values n = 1, 2, . . . , n
co
) is compared against a similarly defined
reference pattern ¯ c[n]. Such comparisons require a distance measure.
For example, the Euclidean distance applied to the cepstrum would give
D =
nco
¸
n=1
[c[n] − ¯ c[n][
2
. (5.19a)
Equivalently in the frequency domain,
D =
1

π
−π

log[H(e
j ˆ ω
)[ − log[
¯
H(e
j ˆ ω
)[

2
dˆ ω, (5.19b)
where log[H(e
j ˆ ω
)[ is the log magnitude of the DTFT of h[n] corre-
sponding to the complex cepstrum in (5.18) or the real part of the
2
For minimum-phase signals, the complex cepstrum satisfies h[n] = 0 for n < 0. Since the
cepstrum is always the even part of the complex cepstrum, it follows that
ˆ
h[n] = 2c[n] for
n > 0.
5.6 Applications to Pattern Recognition 67
DTFT of
ˆ
h[n] in (5.18). Thus, cepstrum-based comparisons are strongly
related to comparisons of smoothed short-time spectra. Therefore, the
cepstrum offers an effective and flexible representation of speech for
pattern recognition problems, and its interpretation as a difference of
log spectra suggests a match to auditory perception mechanisms.
5.6.1 Compensation for Linear Filtering
Suppose that we have only a linearly filtered version of the speech
signal, y[n] = h[n] ∗ x[n], instead of x[n]. If the analysis window is long
compared to the length of h[n], the short-time cepstrum of one frame
of the filtered speech signal y[n] will be
3
c
(y)
ˆ n
[m] = c
(x)
ˆ n
[m] + c
(h)
[m], (5.20)
where c
(h)
[m] will appear essentially the same in each frame. Therefore,
if we can estimate c
(h)
[n],
4
which we assume is non-time-varying, we can
obtain c
(x)
ˆ n
[m] at each frame from c
(y)
ˆ n
[m] by subtraction, i.e., c
(x)
n
[m] =
c
(y)
ˆ n
[m] − c
(h)
[m]. This property is extremely attractive in situations
where the reference pattern ¯ c[m] has been obtained under different
recording or transmission conditions from those used to acquire the
test vector. In these circumstances, the test vector can be compensated
for the effects of the linear filtering prior to computing the distance
measures used for comparison of patterns.
Another approach to removing the effects of linear distortions is to
observe that the cepstrum component due to the distortion is the same
in each frame. Therefore, it can be removed by a simple first difference
operation of the form:
∆c
(y)
ˆ n
[m] = c
(y)
ˆ n
[m] − c
(y)
ˆ n−1
[m]. (5.21)
It is clear that if c
(y)
ˆ n
[m] = c
(x)
ˆ n
[m] + c
(h)
[m] with c
(h)
[m] being indepen-
dent of ˆ n, then ∆c
(y)
ˆ n
[m] = ∆c
(x)
ˆ n
[m], i.e., the linear distortion effects
are removed.
3
In this section, it will be useful to use somewhat more complicated notation. Specifically,
we denote the cepstrum at analysis time ˆ n of a signal x[n] as c
(x)
ˆ n
[m], where m denotes
the quefrency index of the cepstrum.
4
Stockham [124] showed how c
(h)
[m] for such linear distortions can be estimated from the
signal y[n] by time averaging the log of the STFT.
68 Homomorphic Speech Analysis
Furui [39, 40] first noted that the sequence of cepstrum values has
temporal information that could be of value for a speaker verification
system. He used polynomial fits to cepstrum sequences to extract sim-
ple representations of the temporal variation. The delta cepstrum as
defined in (5.21) is simply the slope of a first-order polynomial fit to
the cepstrum time evolution.
5.6.2 Liftered Cepstrum Distance Measures
In using linear predictive analysis (discussed in Chapter 6) to obtain
cepstrum feature vectors for pattern recognition problems, it is
observed that there is significant statistical variability due to a variety
of factors including short-time analysis window position, bias toward
harmonic peaks, and additive noise [63, 126]. A solution to this problem
is to use weighted distance measures of the form:
D =
nco
¸
n=1
g
2
[n] [c[n] − ¯ c[n][
2
, (5.22a)
which can be written as the Euclidean distance of liftered cepstra
D =
nco
¸
n=1
[g[n]c[n] − g[n]¯ c[n][
2
. (5.22b)
Tohkura [126] found, for example, that when averaged over many
frames of speech and speakers, cepstrum values c[n] have zero means
and variances on the order of 1/n
2
. This suggests that g[n] = n for
n = 1, 2, . . . , n
co
could be used to equalize the contributions for each
term to the cepstrum difference.
Juang et al. [63] observed that the variability due to the vagaries of
LPC analysis could be lessened by using a lifter of the form:
g[n] = 1 + 0.5n
co
sin(πn/n
co
) n = 1, 2, . . . , n
co
. (5.23)
Tests of weighted distance measures showed consistent improvements
in automatic speech recognition tasks.
Itakura and Umezaki [55] used the group delay function to derive a
different cepstrum weighting function. Instead of g[n] = n for all n, or
5.6 Applications to Pattern Recognition 69
the lifter of (5.23), Itakura proposed the lifter
g[n] = n
s
e
−n
2
/2τ
2
. (5.24)
This lifter has great flexibility. For example, if s = 0 we have simply
lowpass liftering of the cepstrum. If s = 1 and τ is large, we have essen-
tially g[n] = n for small n with high quefrency tapering. Itakura and
Umezaki [55] tested the group delay spectrum distance measure in an
automatic speech recognition system. They found that for clean test
utterances, the difference in recognition rate was small for different
values of s when τ ≈ 5 although performance suffered with increas-
ing s for larger values of τ. This was attributed to the fact that for
larger s the group delay spectrum becomes very sharply peaked and
thus more sensitive to small differences in formant locations. However,
in test conditions with additive white noise and also with linear filter-
ing distortions, recognition rates improved significantly with τ = 5 and
increasing values of the parameter s.
5.6.3 Mel-Frequency Cepstrum Coefficients
As we have seen, weighted cepstrum distance measures have a directly
equivalent interpretation in terms of distance in the frequency domain.
This is significant in light of models for human perception of sound,
which, as noted in Chapter 3, are based upon a frequency analysis
performed in the inner ear. With this in mind, Davis and Mermelstein
[27] formulated a new type of cepstrum representation that has come to
be widely used and known as the mel-frequency cepstrum coefficients
(mfcc).
The basic idea is to compute a frequency analysis based upon a
filter bank with approximately critical band spacing of the filters and
bandwidths. For 4 kHz bandwidth, approximately 20 filters are used.
In most implementations, a short-time Fourier analysis is done first,
resulting in a DFT X
ˆ n
[k] for analysis time ˆ n. Then the DFT values
are grouped together in critical bands and weighted by a triangular
weighting function as depicted in Figure 5.7. Note that the bandwidths
in Figure 5.7 are constant for center frequencies below 1 kHz and then
increase exponentially up to half the sampling rate of 4 kHz resulting
70 Homomorphic Speech Analysis
0 1000 2000 3000 4000
0
0.005
0.01
frequency in Hz
Fig. 5.7 Weighting functions for Mel-frequency filter bank.
in a total of 22 “filters.” The mel-frequency spectrum at analysis time
ˆ n is defined for r = 1, 2, . . . , R as
MF
ˆ n
[r] =
1
A
r
Ur
¸
k=Lr
[V
r
[k]X
ˆ n
[k][
2
, (5.25a)
where V
r
[k] is the triangular weighting function for the rth filter ranging
from DFT index L
r
to U
r
, where
A
r
=
Ur
¸
k=Lr
[V
r
[k][
2
(5.25b)
is a normalizing factor for the rth mel-filter. This normalization is built
into the weighting functions of Figure 5.7. It is needed so that a per-
fectly flat input Fourier spectrum will produce a flat mel-spectrum. For
each frame, a discrete cosine transform of the log of the magnitude of
the filter outputs is computed to form the function mfcc
ˆ n
[m], i.e.,
mfcc
ˆ n
[m] =
1
R
R
¸
r=1
log(MF
ˆ n
[r])cos
¸

R

r +
1
2

m

. (5.26)
Typically, mfcc
ˆ n
[m] is evaluated for a number of coefficients, N
mfcc
,
that is less than the number of mel-filters, e.g., N
mfcc
= 13 and R = 22.
Figure 5.8 shows the result of mfcc analysis of a frame of voiced speech
in comparison with the short-time Fourier spectrum, LPC spectrum
(discussed in Chapter 6), and a homomorphically smoothed spectrum.
The large dots are the values of log(MF
ˆ n
[r]) and the line interpolated
5.7 The Role of the Cepstrum 71
0 500 1000 1500 2000 2500 3000 3500 4000
−7
−6
−5
−4
−3
−2
−1
0
1
2
3
l
o
g
m
a
g
n
i
t
u
d
e
frequency in Hz
Short-time Fourier Transform
Homomorphic smoothing,
LPC smoothing,
Mel cepstrum smoothing,
Fig. 5.8 Comparison of spectral smoothing methods.
between them is a spectrum reconstructed at the original DFT frequen-
cies. Note that all these spectra are different, but they have in common
that they have peaks at the formant resonances. At higher frequencies,
the reconstructed mel-spectrum of course has more smoothing due to
the structure of the filter bank.
Note that the delta cepstrum idea expressed by (5.21) can be applied
to mfcc to remove the effects of linear filtering as long as the frequency
response of the distorting linear filter does not vary much across each
of the mel-frequency bands.
5.7 The Role of the Cepstrum
As discussed in this chapter, the cepstrum entered the realm of speech
processing as a basis for pitch detection. It remains one of the most
effective indicators of voice pitch that have been devised. Because the
72 Homomorphic Speech Analysis
vocal tract and excitation components are well separated in the cep-
strum, it was natural to consider analysis techniques for estimation of
the vocal tract system as well [86, 88, 111]. While separation techniques
based on the cepstrum can be very effective, the linear predictive anal-
ysis methods to be discussed in the next chapter have proven to be
more effective for a variety of reasons. Nevertheless, the cepstrum con-
cept has demonstrated its value when applied to vocal tract system
estimates obtained by linear predictive analysis.
6
Linear Predictive Analysis
Linear predictive analysis is one of the most powerful and widely
used speech analysis techniques. The importance of this method lies
both in its ability to provide accurate estimates of the speech param-
eters and in its relative speed of computation. In this chapter, we
present a formulation of the ideas behind linear prediction, and dis-
cuss some of the issues that are involved in using it in practical speech
applications.
6.1 Linear Prediction and the Speech Model
We come to the idea of linear prediction of speech by recalling the
source/system model that was introduced in Chapter 2, where the
sampled speech signal was modeled as the output of a linear, slowly
time-varying system excited by either quasi-periodic impulses (during
voiced speech), or random noise (during unvoiced speech). The par-
ticular form of the source/system model implied by linear predictive
analysis is depicted in Figure 6.1, where the speech model is the part
inside the dashed box. Over short time intervals, the linear system is
73
74 Linear Predictive Analysis
Fig. 6.1 Model for linear predictive analysis of speech signals.
described by an all-pole system function of the form:
H(z) =
S(z)
E(z)
=
G
1 −
p
¸
k=1
a
k
z
−k
. (6.1)
In linear predictive analysis, the excitation is defined implicitly by the
vocal tract system model, i.e., the excitation is whatever is needed
to produce s[n] at the output of the system. The major advantage of
this model is that the gain parameter, G, and the filter coefficients
¦a
k
¦ can be estimated in a very straightforward and computationally
efficient manner by the method of linear predictive analysis.
For the system of Figure 6.1 with the vocal tract model of (6.1), the
speech samples s[n] are related to the excitation e[n] by the difference
equation
s[n] =
p
¸
k=1
a
k
s[n − k] + Ge[n]. (6.2)
A linear predictor with prediction coefficients, α
k
, is defined as a system
whose output is
˜ s[n] =
p
¸
k=1
α
k
s[n − k], (6.3)
and the prediction error, defined as the amount by which ˜ s[n] fails to
exactly predict sample s[n], is
d[n] = s[n] − ˜ s[n] = s[n] −
p
¸
k=1
α
k
s[n − k]. (6.4)
6.1 Linear Prediction and the Speech Model 75
From (6.4) it follows that the prediction error sequence is the output
of an FIR linear system whose system function is
A(z) = 1 −
p
¸
k=1
α
k
z
−k
=
D(z)
S(z)
. (6.5)
It can be seen by comparing (6.2) and (6.4) that if the speech signal
obeys the model of (6.2) exactly, and if α
k
= a
k
, then d[n] = Ge[n].
Thus, the prediction error filter, A(z), will be an inverse filter for the
system, H(z), of (6.1), i.e.,
H(z) =
G
A(z)
. (6.6)
The basic problem of linear prediction analysis is to determine the
set of predictor coefficients ¦α
k
¦ directly from the speech signal in
order to obtain a useful estimate of the time-varying vocal tract system
through the use of (6.6). The basic approach is to find a set of predic-
tor coefficients that will minimize the mean-squared prediction error
over a short segment of the speech waveform. The resulting param-
eters are then assumed to be the parameters of the system function
H(z) in the model for production of the given segment of the speech
waveform. This process is repeated periodically at a rate appropriate
to track the phonetic variation of speech (i.e., order of 50–100 times
per second).
That this approach will lead to useful results may not be immedi-
ately obvious, but it can be justified in several ways. First, recall that
if α
k
= a
k
, then d[n] = Ge[n]. For voiced speech this means that d[n]
would consist of a train of impulses, i.e., d[n] would be small except at
isolated samples spaced by the current pitch period, P
0
. Thus, finding
α
k
s that minimize the mean-squared prediction error seems consistent
with this observation. A second motivation for this approach follows
from the fact that if a signal is generated by (6.2) with non-time-varying
coefficients and excited either by a single impulse or by a stationary
white noise input, then it can be shown that the predictor coefficients
that result from minimizing the mean-squared prediction error (over
all time) are identical to the coefficients of (6.2). A third pragmatic
justification for using the minimum mean-squared prediction error as a
76 Linear Predictive Analysis
basis for estimating the model parameters is that this approach leads to
an exceedingly useful and accurate representation of the speech signal
that can be obtained by efficient solution of a set of linear equations.
The short-time average prediction error is defined as
E
ˆ n
=

d
2
ˆ n
[m]

=

s
ˆ n
[m] −
p
¸
k=1
α
k
s
ˆ n
[m − k]

2
¸
, (6.7)
where s
ˆ n
[m] is a segment of speech that has been selected in a neigh-
borhood of the analysis time ˆ n, i.e.,
s
ˆ n
[m] = s[m + ˆ n] − M
1
≤ m≤ M
2
. (6.8)
That is, the time origin of the analysis segment is shifted to sample
ˆ n of the entire signal. The notation ' ` denotes averaging over a finite
number of samples. The details of specific definitions of the averaging
operation will be discussed below.
We can find the values of α
k
that minimize E
ˆ n
in (6.7) by setting
∂E
ˆ n
/∂α
i
= 0, for i = 1, 2, . . . , p, thereby obtaining the equations
1
p
¸
k=1
˜ α
k
's
ˆ n
[m − i]s
ˆ n
[m − k]` = 's
ˆ n
[m − i]s
ˆ n
[m]` 1 ≤ i ≤ p, (6.9)
where the ˜ α
k
are the values of α
k
that minimize E
ˆ n
in (6.7). (Since the
˜ α
k
are unique, we will drop the tilde and use the notation α
k
to denote
the values that minimize E
ˆ n
.) If we define
ϕ
ˆ n
[i, k] = 's
ˆ n
[m − i]s
ˆ n
[m − k]`, (6.10)
then (6.9) can be written more compactly as
2
p
¸
k=1
α
k
ϕ
ˆ n
[i, k] = ϕ
ˆ n
[i, 0] i = 1, 2, . . . , p. (6.11)
1
More accurately, the solutions of (6.7) only provide a stationary point which can be shown
to be a minimum of E
ˆ n
since E
ˆ n
is a convex function of the α
i
s.
2
The quantities ϕ
ˆ n
[i, k] are in the form of a correlation function for the speech segment
s
ˆ n
[m]. The details of the definition of the averaging operation used in (6.10) have a
significant effect on the properties of the prediction coefficients that are obtained by solving
(6.11).
6.2 Computing the Prediction Coefficients 77
If we know ϕ
ˆ n
[i, k] for 1 ≤ i ≤ p and 0 ≤ k ≤ p, this set of p equations
in p unknowns, which can be represented by the matrix equation:
Φα = ψ, (6.12)
can be solved for the vector α = ¦α
k
¦ of unknown predictor coefficients
that minimize the average squared prediction error for the segment
s
ˆ n
[m].
3
Using (6.7) and (6.9), the minimum mean-squared prediction
error can be shown to be [3, 5]
E
ˆ n
= ϕ
ˆ n
[0, 0] −
p
¸
k=1
α
k
ϕ
ˆ n
[0, k]. (6.13)
Thus, the total minimum mean-squared error consists of a fixed com-
ponent equal to the mean-squared value of the signal segment minus a
term that depends on the predictor coefficients that satisfy (6.11), i.e.,
the optimum coefficients reduce E
ˆ n
in (6.13) the most.
To solve for the optimum predictor coefficients, we must first com-
pute the quantities ϕ
ˆ n
[i, k] for 1 ≤ i ≤ p and 0 ≤ k ≤ p. Once this
is done we only have to solve (6.11) to obtain the α
k
s. Thus, in
principle, linear prediction analysis is very straightforward. However,
the details of the computation of ϕ
ˆ n
[i, k] and the subsequent solu-
tion of the equations are somewhat intricate and further discussion
is required.
6.2 Computing the Prediction Coefficients
So far we have not been explicit about the meaning of the averaging
notation ' ` used to define the mean-squared prediction error in (6.7).
As we have stated, in a short-time analysis procedure, the averaging
must be over a finite interval. We shall see below that two methods for
linear predictive analysis emerge out of a consideration of the limits of
summation and the definition of the waveform segment s
ˆ n
[m].
4
3
Although the α
k
s are functions of ˆ n (the time index at which they are estimated) this
dependence will not be explicitly shown. We shall also find it advantageous to drop the
subscripts ˆ n on E
ˆ n
, s
ˆ n
[m], and ϕ
ˆ n
[i, k] when no confusion will result.
4
These two methods, applied to the same speech signal yield slightly different optimum
predictor coefficients.
78 Linear Predictive Analysis
6.2.1 The Covariance Method
One approach to computing the prediction coefficients is based on the
definition
E
ˆ n
=
M
2
¸
m=−M
1
(d
ˆ n
[m])
2
=
M
2
¸
m=−M
1

s
ˆ n
[m] −
p
¸
k=1
α
k
s
ˆ n
[m − k]

2
, (6.14)
where −M
1
≤ n ≤ M
2
. The quantities ϕ
ˆ n
[i, k] needed in (6.11) inherit
the same definition of the averaging operator, i.e.,
ϕ
ˆ n
[i, k] =
M
2
¸
m=−M
1
s
ˆ n
[m − i]s
ˆ n
[m − k]

1 ≤ i ≤ p
0 ≤ k ≤ p.
(6.15)
Both (6.14) and (6.15) require values of s
ˆ n
[m] = s[m + ˆ n] over the
range −M
1
− p ≤ m≤ M
2
. By changes of index of summation, (6.15)
can be expressed in the equivalent forms:
ϕ
ˆ n
[i, k] =
M
2
−i
¸
m=−M
1
−i
s
ˆ n
[m]s
ˆ n
[m + i − k] (6.16a)
=
M
2
−k
¸
m=−M
1
−k
s
ˆ n
[m]s
ˆ n
[m + k − i], (6.16b)
from which it follows that ϕ
ˆ n
[i, k] = ϕ
ˆ n
[k, i].
Figure 6.2 shows the sequences that are involved in computing the
mean-squared prediction error as defined by (6.14). The top part of
this figure shows a sampled speech signal s[m], and the box denotes
a segment of that waveform selected around some time index ˆ n. The
second plot shows that segment extracted as a finite-length sequence
s
ˆ n
[m] = s[m + ˆ n] for −M
1
− p ≤ m≤ M
2
. Note the p “extra” samples
(light shading) at the beginning that are needed to start the prediction
error filter at time −M
1
. This method does not require any assump-
tion about the signal outside the interval −M
1
− p ≤ m≤ M
2
since
the samples s
ˆ n
[m] for −M
1
− p ≤ m≤ M
2
are sufficient to evaluate
ϕ[i, k] for all required values of i and k. The third plot shows the pre-
diction error computed with the optimum predictor coefficients. Note
that in solving for the optimum prediction coefficients using (6.11), the
6.2 Computing the Prediction Coefficients 79
Fig. 6.2 Illustration of windowing and prediction error for the covariance method.
prediction error is implicitly computed over the range −M
1
≤ m≤ M
2
as required in (6.14) and (6.15). The minimum mean-squared predic-
tion error would be simply the sum of squares of all the samples shown.
In most cases of linear prediction, the prediction error sequence is not
explicitly computed since solution of (6.11) does not require it.
The mathematical structure that defines the covariance method of
linear predictive analysis implies a number of useful properties of the
solution. It is worthwhile to summarize them as follows:
(C.1) The mean-squared prediction error satisfies E
ˆ n
≥ 0. With
this method, it is theoretically possible for the average
error to be exactly zero.
(C.2) The matrix Φ in (6.12) is a symmetric positive-semi-
definite matrix.
(C.3) The roots of the prediction error filter A(z) in (6.5) are
not guaranteed to lie within the unit circle of the z-plane.
This implies that the vocal tract model filter (6.6) is not
guaranteed to be stable.
80 Linear Predictive Analysis
(C.4) As a result of (C.2), the equations (6.11) can be solved
efficiently using, for example, the well known Cholesky
decomposition of the covariance matrix Φ [3].
6.2.2 The Autocorrelation Method
Perhaps the most widely used method of linear predictive analysis
is called the autocorrelation method because the covariance function
ϕ
ˆ n
[i, k] needed in (6.11) reduces to the STACF φ
ˆ n
[[i − k[] that we dis-
cussed in Chapter 4 [53, 54, 74, 78]. In the autocorrelation method, the
analysis segment s
ˆ n
[m] is defined as
s
ˆ n
[m] =

s[n + m]w[m] −M
1
≤ m≤ M
2
0 otherwise,
(6.17)
where the analysis window w[m] is used to taper the edges of the seg-
ment to zero. Since the analysis segment is defined by the windowing
of (6.17) to be zero outside the interval −M
1
≤ m≤ M
2
, it follows that
the prediction error sequence d
ˆ n
[m] can be nonzero only in the range
−M
1
≤ m≤ M
2
+ p. Therefore, E
ˆ n
is defined as
E
ˆ n
=
M
2
+p
¸
m=−M
1
(d
ˆ n
[m])
2
=

¸
m=−∞
(d
ˆ n
[m])
2
. (6.18)
The windowing of (6.17) allows us to use the infinite limits to signify
that the sum is over all nonzero values of d
ˆ n
[m]. Applying this notion
to (6.16a) and (6.16b) leads to the conclusion that
ϕ
ˆ n
[i, k] =

¸
m=−∞
s
ˆ n
[m]s
ˆ n
[m + [i − k[] = φ
ˆ n
[[i − k[]. (6.19)
Thus, ϕ[i, k] is a function only of [i − k[. Therefore, we can replace
ϕ
ˆ n
[i, k] by φ
ˆ n
[[i − k[], which is the STACF defined in Chapter 4 as
φ
ˆ n
[k] =

¸
m=−∞
s
ˆ n
[m]s
ˆ n
[m + k] = φ
ˆ n
[−k]. (6.20)
The resulting set of equations for the optimum predictor coefficients is
therefore
p
¸
k=1
α
k
φ
ˆ n
[[i − k[] = φ
ˆ n
[i] i = 1, 2, . . . , p. (6.21)
6.2 Computing the Prediction Coefficients 81
Fig. 6.3 Illustration of windowing and prediction error for the autocorrelation method.
Figure 6.3 shows the sequences that are involved in computing the
optimum prediction coefficients using the autocorrelation method. The
upper plot shows the same sampled speech signal s[m] as in Figure 6.2
with a Hamming window centered at time index ˆ n. The middle plot
shows the result of multiplying the signal s[ˆ n + m] by the window w[m]
and redefining the time origin to obtain s
ˆ n
[m]. Note that the zero-
valued samples outside the window are shown with light shading. The
third plot shows the prediction error computed using the optimum
coefficients. Note that for this segment, the prediction error (which is
implicit in the solution of (6.21)) is nonzero over the range −M
1
≤ m≤
M
2
+ p. Also note the lightly shaded p samples at the beginning. These
samples can be large due to the fact that the predictor must predict
these samples from the zero-valued samples that precede the windowed
segment s
ˆ n
[m]. It is easy to see that at least one of these first p samples
of the prediction error must be nonzero. Similarly, the last p samples of
the prediction error can be large due to the fact that the predictor must
predict zero-valued samples from windowed speech samples. It can also
be seen that at least one of these last p samples of the prediction error
must be nonzero. For this reason, it follows that E
ˆ n
, being the sum of
82 Linear Predictive Analysis
squares of the prediction error samples, must always be strictly greater
than zero.
5
As in the case of the covariance method, the mathematical structure
of the autocorrelation method implies a number of properties of the
solution, including the following:
(A.1) The mean-squared prediction error satisfies E
ˆ n
> 0. With
this method, it is theoretically impossible for the error
to be exactly zero because there will always be at least
one sample at the beginning and one at the end of the
prediction error sequence that will be nonzero.
(A.2) The matrix Φ in (6.12) is a symmetric positive-definite
Toeplitz matrix [46].
(A.3) The roots of the prediction error filter A(z) in (6.5) are
guaranteed to lie within the unit circle of the z-plane so
that the vocal tract model filter of (6.6) is guaranteed to
be stable.
(A.4) As a result of (A.2), the equations (6.11) can be solved
efficiently using the Levinson–Durbin algorithm, which,
because of its many implications, we discuss in more
detail in Section 6.3.
6.3 The Levinson–Durbin Recursion
As stated in (A.2) above, the matrix Φin (6.12) is a symmetric positive-
definite Toeplitz matrix, which means that all the elements on a given
diagonal in the matrix are equal. Equation (6.22) shows the detailed
structure of the matrix equation Φα = ψ for the autocorrelation
method.

φ[0] φ[1] φ[p − 1]
φ[1] φ[0] φ[p − 2]

φ[p − 1] φ[p − 2] φ[0]
¸
¸
¸
¸
¸

α
1
α
2

α
p
¸
¸
¸
¸
¸
=

φ[1]
φ[2]

φ[p]
¸
¸
¸
¸
¸
(6.22)
5
For this reason, a tapering window is generally used in the autocorrelation method.
6.3 The Levinson–Durbin Recursion 83
Note that the vector ψ is composed of almost the same autocorrelation
values as comprise Φ. Because of the special structure of (6.22) it is
possible to derive a recursive algorithm for inverting the matrix Φ.
That algorithm, known as the Levinson–Durbin algorithm, is specified
by the following steps:
Levinson–Durbin Algorithm
E
0
= φ[0] (D.1)
for i = 1, 2, . . . , p
k
i
=

¸
φ[i] −
i−1
¸
j=1
α
(i−1)
j
φ[i − j]
¸

E
(i−1)
(D.2)
α
(i)
i
= k
i
(D.3)
if i > 1 then for j = 1, 2, . . . , i − 1
α
(i)
j
= α
(i−1)
j
− k
i
α
(i−1)
i−j
(D.4)
end
E
(i)
= (1 − k
2
i
)E
(i−1)
(D.5)
end
α
j
= α
(p)
j
j = 1, 2, . . . , p (D.6)
An important feature of the Levinson–Durbin algorithm is that it
determines by recursion the optimum ith-order predictor from the opti-
mum (i − 1)th-order predictor, and as part of the process, all predic-
tors from order 0 (no prediction) to order p are computed along with
the corresponding mean-squared prediction errors E
(i)
. Specifically, the
equations labeled (D.3) and (D.4) can be used to show that the pre-
diction error system function satisfies
A
(i)
(z) = A
(i−1)
(z) − k
i
z
−i
A
(i−1)
(z
−1
). (6.23)
Defining the ith-order forward prediction error e
(i)
[n] as the out-
put of the prediction error filter with system function A
(i)
(z) and
b
(i)
[n] as the output of the ith-order backward prediction error filter
B
(i)
(z) = z
−i
A
(i)
(z
−1
), (6.23) leads (after some manipulations) to an
interpretation of the Levinson–Durbin algorithm in terms of a lattice
filter structure as in Figure 6.4(a).
84 Linear Predictive Analysis
Fig. 6.4 Lattice structures derived from the Levinson–Durbin recursion. (a) Prediction error
filter A(z). (b) Vocal tract filter H(z) = 1/A(z).
Also, by solving two equations in two unknowns recursively, it is
possible to start at the output of Figure 6.4(a) and work to the left
eventually computing s[n] in terms of e[n]. The lattice structure corre-
sponding to this is shown in Figure 6.4(b). Lattice structures like this
can be derived from acoustic principles applied to a physical model
composed of concatenated lossless tubes [101]. If the input and output
signals in such physical models are sampled at just the right sampling
rate, the sampled signals are related by a transfer function identical to
(6.6). In this case, the coefficients k
i
behave as reflection coefficients at
the tube boundaries [3, 78, 101].
Note that the parameters k
i
for i = 1, 2, . . . , p play a key role in
the Levinson–Durbin recursion and also in the lattice filter interpreta-
tion. Itakura and Saito [53, 54], showed that the parameters k
i
in the
Levinson–Durbin recursion and the lattice filter interpretation obtained
from it also could be derived by looking at linear predictive analysis
from a statistical perspective. They called the k
i
parameters, PARCOR
(for partial correlation) coefficients [54], because they can be computed
directly as a ratio of cross-correlation values between the forward and
6.4 LPC Spectrum 85
backward prediction errors at the output of the (i − 1)th stage of pre-
diction in Figure 6.4(a), i.e.,
k
i
=

¸
m=−∞
e
(i−1)
[m]b
(i−1)
[m − 1]


¸
m=−∞

e
(i−1)
[m]

2

¸
m=−∞

b
(i−1)
[m − 1]

2

1/2
. (6.24)
In the PARCOR interpretation, each stage of Figure 6.4(a) removes
part of the correlation in the input signal. The PARCOR coefficients
computed using (6.24) are identical to the k
i
obtained as a result of the
Levinson–Durbin algorithm. Indeed, Equation (D.2) in the Levinson–
Durbin algorithm can be replaced by (6.24), and the result is an algo-
rithm for transforming the PARCOR representation into the linear pre-
dictor coefficient representation.
The Levinson–Durbin formulation provides one more piece of useful
information about the PARCOR coefficients. Specifically, from Equa-
tion (D.5) of the algorithm, it follows that, since E
(i)
= (1 − k
2
i
)E
(i−1)
is strictly greater than zero for predictors of all orders, it must be true
that −1 < k
i
< 1 for all i. It can be shown that this condition on the
PARCORs also guarantees that all the zeros of a prediction error filter
A
(i)
(z) of any order must be strictly inside the unit circle of the z-plane
[74, 78].
6.4 LPC Spectrum
The frequency-domain interpretation of linear predictive analysis pro-
vides an informative link to our earlier discussions of the STFT and cep-
strum analysis. The autocorrelation method is based on the short-time
autocorrelation function, φ
ˆ n
[m], which is the inverse discrete Fourier
transform of the magnitude-squared of the STFT, [S
ˆ n
(e
j ˆ ω
)[
2
, of the
windowed speech signal s
ˆ n
[m] = s[n + m]w[m]. The values φ
ˆ n
[m] for
m = 0, 1, . . . , p are used to compute the prediction coefficients and gain,
which in turn define the vocal tract system function H(z) in (6.6).
Therefore, the magnitude-squared of the frequency response of this sys-
tem, obtained by evaluating H(z) on the unit circle at angles 2πf/f
s
,
86 Linear Predictive Analysis
is of the form:
[H(e
j2πf/fs
)[
2
=

G
1 −
p
¸
k=1
α
k
e
−j2πf/fs

2
, (6.25)
and can be thought of as an alternative short-time spectral representa-
tion. Figure 6.5 shows a comparison between short-time Fourier analy-
sis and linear predictive spectrum analysis for segments of voiced and
unvoiced speech. Figures 6.5(a) and 6.5(c) show the STACF, with the
first 23 values plotted with a heavy line. These values are used to
compute the predictor coefficients and gain for an LPC model with
p = 22. The frequency responses of the corresponding vocal tract sys-
tem models are computed using (6.25) where the sampling frequency is
f
s
= 16 kHz. These frequency responses are superimposed on the corre-
sponding STFTs (shown in gray). As we have observed before, the rapid
variations with frequency in the STFT are due primarily to the exci-
tation, while the overall shape is assumed to be determined by the
effects of glottal pulse, vocal tract transfer function, and radiation. In
0 5 10 15
−0.5
0
0.5
1
0 5 10 15
−0.5
0
0.5
1
0 2000 4000 6000 8000
−100
−50
0
50
0 2000 4000 6000 8000
−100
−50
0
50
(b) Voiced LPC Spectrum (a) Voiced Autocorrelation
(d) Unvoiced LPC Spectrum (c) Unvoiced Autocorrelation
time in ms frequency in kHz
Fig. 6.5 Comparison of short-time Fourier analysis with linear predictive analysis.
6.4 LPC Spectrum 87
cepstrum analysis, the excitation effects are removed by lowpass lifter-
ing the cepstrum. In linear predictive spectrum analysis, the excitation
effects are removed by focusing on the low-time autocorrelation coef-
ficients. The amount of smoothing of the spectrum is controlled by
the choice of p. Figure 6.5 shows that a linear prediction model with
p = 22 matches the general shape of the short-time spectrum, but does
not represent all its local peaks and valleys, and this is exactly what is
desired.
The question naturally arises as to how p should be chosen.
Figure 6.6 offers a suggestion. In Figure 6.6(a) are shown the STFT
(in gray) and the frequency responses of a 12th-order model (heavy
dark line) and a 40th-order model (thin dark line). Evidently, the lin-
ear predictive spectra tend to favor the peaks of the short-time Fourier
transform. That this is true in general can be argued using the Parseval
theorem of Fourier analysis [74]. This is in contrast to homomorphic
Fig. 6.6 Effect of predictor order on: (a) estimating the spectral envelope of a 4-kHz band-
width signal; and (b) the normalized mean-squared prediction error for the same signal.
88 Linear Predictive Analysis
smoothing of the STFT, which tends toward an average of the peaks
and valleys of the STFT. As an example, see Figure 5.8, which shows
a comparison between the short-time Fourier spectrum, the LPC spec-
trum with p = 12, a homomorphically smoothed spectrum using 13
cepstrum values, and a mel-frequency spectrum.
Figure 6.6(b) shows the normalized mean-squared prediction error
V
(p)
= E
(p)
/φ[0] as a function of p for the segment of speech used to
produce Figure 6.6(a). Note the sharp decrease from p = 0 (no pre-
diction implies V
(0)
= 1) to p = 1 and the less abrupt decrease there-
after. Furthermore, notice that the mean-squared error curve flattens
out above about p = 12 and then decreases modestly thereafter. The
choice p = 12 gives a good match to the general shape of the STFT,
highlighting the formant structure imposed by the vocal tract filter
while ignoring the periodic pitch structure. Observe that p = 40 gives
a much different result. In this case, the vocal tract model is highly
influenced by the pitch harmonics in the short-time spectrum. It can
be shown that if p is increased to the length of the windowed speech
segment, that [H(e
j2πf/fs
)[
2
→[S
ˆ n
(e
j2πf/fs
)[
2
, but as pointed out in
(A.1) above, E
(2M)
does not go to zero [75].
In a particular application, the prediction order is generally fixed
at a value that captures the general spectral shape due to the glot-
tal pulse, vocal tract resonances, and radiation. From the acoustic
theory of speech production, it follows that the glottal pulse spec-
trum is lowpass, the radiation filtering is highpass, and the vocal tract
imposes a resonance structure that, for adult speakers, is comprised
of about one resonance per kilohertz of frequency [101]. For the sam-
pled speech signal, the combination of the lowpass glottal pulse spec-
trum and the highpass filtering of radiation are usually adequately
represented by one or two additional complex pole pairs. When cou-
pled with an estimate of one resonance per kilohertz, this leads to a
rule of thumb of p = 4 + f
s
/1000. For example, for a sampling rate of
f
s
= 8000 Hz, it is common to use a predictor order of 10–12. Note that
in Figure 6.5 where the sampling rate was f
s
= 16000 Hz, a predictor
order of p = 22 gave a good representation of the overall shape and
resonance structure of the speech segment over the band from 0 to
8000 Hz.
6.5 Equivalent Representations 89
The example of Figure 6.6 illustrates an important point about
linear predictive analysis. In that example, the speaker was a high-
pitched female, and the wide spacing between harmonics causes the
peaks of the linear predictive vocal tract model to be biased toward
those harmonics. This effect is a serious limitation for high-pitched
voices, but is less problematic for male voices where the spacing between
harmonics is generally much smaller.
6.5 Equivalent Representations
The basic parameters obtained by linear predictive analysis are the gain
G and the prediction coefficients ¦α
k
¦. From these, a variety of different
equivalent representations can be obtained. These different representa-
tions are important, particularly when they are used in speech coding,
because we often wish to quantize the model parameters for efficiency
in storage or transmission of coded speech.
In this section, we give only a brief summary of the most important
equivalent representations.
6.5.1 Roots of Prediction Error System Function
Equation (6.5) shows that the system function of the prediction error
filter is a polynomial in z
−1
and therefore it can be represented in terms
of its zeros as
A(z) = 1 −
p
¸
k=1
α
k
z
−k
=
p
¸
k=1
(1 − z
k
z
−1
). (6.26)
According to (6.6), the zeros of A(z) are the poles of H(z). Therefore,
if the model order is chosen judiciously as discussed in the previous
section, then it can be expected that roughly f
s
/1000 of the roots will
be close in frequency (angle in the z-plane) to the formant frequencies.
Figure 6.7 shows an example of the roots (marked with ) of a 12th-
order predictor. Note that eight (four complex conjugate pairs) of the
roots are close to the unit circle. These are the poles of H(z) that model
the formant resonances. The remaining four roots lie well within the
unit circle, which means that they only provide for the overall spectral
shaping resulting from the glottal and radiation spectral shaping.
90 Linear Predictive Analysis
Fig. 6.7 Poles of H(z) (zeros of A(z)) marked with × and LSP roots marked with ∗ and o.
The prediction coefficients are perhaps the most susceptible to quan-
tization errors of all the equivalent representations. It is well-known
that the roots of a polynomial are highly sensitive to errors in its coef-
ficients — all the roots being a function of all the coefficients [89].
Since the pole locations are crucial to accurate representation of the
spectrum, it is important to maintain high accuracy in the location of
the zeros of A(z). One possibility, not often invoked, would be to factor
the polynomial as in (6.26) and then quantize each root (magnitude
and angle) individually.
6.5.2 LSP Coefficients
A much more desirable alternative to quantization of the roots of A(z)
was introduced by Itakura [52], who defined the line spectrum pair
(LSP) polynomials
6
P(z) = A(z) + z
−(p+1)
A(z
−1
) (6.27a)
Q(z) = A(z) − z
−(p+1)
A(z
−1
). (6.27b)
6
Note the similarity to (6.23), from which it follows that P(z) and Q(z) are system functions
of lattice filters obtained by extending Figure 6.4(a) with an additional section with k
p+1
=
∓1, respectively.
6.5 Equivalent Representations 91
This transformation of the linear prediction parameters is invertible:
to recover A(z) from P(z) and Q(z) simply add the two equations to
obtain A(z) = (P(z) + Q(z))/2.
An illustration of the LSP representation is given in Figure 6.7,
which shows the roots of A(z), P(z), and Q(z) for a 12th-order pre-
dictor. This new representation has some very interesting and useful
properties that are confirmed by Figure 6.7 and summarized below
[52, 119].
(LSP.1) All the roots of P(z) and Q(z) are on the unit circle,
i.e., they can be represented as
P(z) =
p+1
¸
k=1
(1 − e
jΩ
k
z
−1
) (6.28a)
Q(z) =
p+1
¸
k=1
(1 − e

k
z
−1
). (6.28b)
The normalized discrete-time frequencies (angles in the
z-plane), Ω
k
and Θ
k
, are called the line spectrum fre-
quencies or LSFs. Knowing the p LSFs that lie in the
range 0 < ω < π is sufficient to completely define the
LSP polynomials and therefore A(z).
(LSP.2) If p is an even integer, then P(−1) = 0 and Q(1) = 0.
(LSP.3) The PARCOR coefficients corresponding to A(z) sat-
isfy [k
i
[ < 1 if and only if the roots of P(z) and Q(z)
alternate on the unit circle, i.e., the LSFs are interlaced
over the range of angles 0 to π.
(LSP.4) The LSFs are close together when the roots of A(z)
are close to the unit circle.
These properties make it possible to represent the linear predictor
by quantized differences between the successive LSFs [119]. If prop-
erty (LSP.3) is maintained through the quantization, the reconstructed
polynomial A(z) will still have its zeros inside the unit circle.
92 Linear Predictive Analysis
6.5.3 Cepstrum of Vocal Tract Impulse Response
One of the most useful alternative representations of the linear predic-
tor is the cepstrum of the impulse response, h[n], of the vocal tract
filter. The impulse response can be computed recursively through the
difference equation:
h[n] =
p
¸
k=1
α
k
h[n − k] + Gδ[n], (6.29)
or a closed form expression for the impulse response can be obtained
by making a partial fraction expansion of the system function H(z).
As discussed in Section 5.3.3, since H(z) is a minimum-phase sys-
tem (all its poles inside the unit circle), it follows that the complex cep-
strum of h[n] can be computed using (5.14), which would require that
the poles of H(z) be found by polynomial rooting. Then, any desired
number of cepstrum values can be computed. However, because of the
minimum phase condition, a more direct recursive computation is pos-
sible [101]. Equation (5.16) gives the recursion for computing
ˆ
h[n] from
the predictor coefficients and gain, and (5.17) gives the inverse recursion
for computing the predictor coefficients from the complex cepstrum of
the vocal tract impulse response. These relationships are particularly
useful in speech recognition applications where a small number of cep-
strum values is used as a feature vector.
6.5.4 PARCOR Coefficients
We have seen that the PARCOR coefficients are bounded by ±1. This
makes them an attractive parameter for quantization.
We have mentioned that if we have a set of PARCOR coefficients,
we can simply use them at step (D.2) of the Levinson–Durbin algo-
rithm thereby obtaining an algorithm for converting PARCORs to
predictor coefficients. By working backward through the Levinson–
Durbin algorithm, we can compute the PARCOR coefficients from
a given set of predictor coefficients. The resulting algorithm is
given below.
6.5 Equivalent Representations 93
Predictor-to-PARCOR Algorithm
α
(p)
j
= α
j
j = 1, 2, . . . , p
k
p
= α
(p)
p
(P.1)
for i = p, p − 1, . . . , 2
for j = 1, 2, . . . , i − 1
α
(i−1)
j
=
α
(i)
j
+ k
i
α
(i)
i−j
1 − k
2
i
(P.2)
end
k
i−1
= α
(i−1)
i−1
(P.3)
end
6.5.5 Log Area Coefficients
As mentioned above, the lattice filter representation of the vocal tract
system is strongly related to concatenated acoustic tube models of the
physics of sound propagation in the vocal tract. Such models are char-
acterized by a set of tube cross-sectional areas denoted A
i
. These “area
function” parameters are useful alternative representations of the vocal
tract model obtained from linear predictive analysis. Specifically, the
log area ratio parameters are defined as
g
i
= log
¸
A
i+1
A
i

= log
¸
1 − k
i
1 + k
i

, (6.30)
where A
i
and A
i+1
are the areas of two successive tubes and the PAR-
COR coefficient −k
i
is the reflection coefficient for sound waves imping-
ing on the junction between the two tubes [3, 78, 101]. The inverse
transformation (from g
i
to k
i
) is
k
i
=
1 − e
g
i
1 + e
g
i
. (6.31)
The PARCOR coefficients can be converted to predictor coefficients if
required using the technique discussed in Section 6.5.4.
Viswanathan and Makhoul [131] showed that the frequency response
of a vocal tract filter represented by quantized log area ratio coefficients
is relatively insensitive to quantization errors.
94 Linear Predictive Analysis
6.6 The Role of Linear Prediction
In this chapter, we have discussed many of the fundamentals of linear
predictive analysis of speech. The value of linear predictive analysis
stems from its ability to represent in a compact form the part of the
speech model associated with the vocal tract, which in turn is closely
related to the phonemic representation of the speech signal. Over 40
years of research on linear predictive methods have yielded a wealth of
knowledge that has been widely and effectively applied in almost every
area of speech processing, but especially in speech coding, speech syn-
thesis, and speech recognition. The present chapter and Chapters 3–5
provide the basis for the remaining chapters of this text, which focus
on these particular application areas.
7
Digital Speech Coding
This chapter, the first of three on digital speech processing applications,
focuses on specific techniques that are used in digital speech coding.
We begin by describing the basic operation of sampling a speech signal
and directly quantizing and encoding of the samples. The remainder
of the chapter discusses a wide variety of techniques that represent the
speech signal in terms of parametric models of speech production and
perception.
7.1 Sampling and Quantization of Speech (PCM)
In any application of digital signal processing (DSP), the first step is
sampling and quantization of the resulting samples into digital form.
These operations, which comprise the process of A-to-D conversion,
are depicted for convenience in analysis and discussion in the block
diagram of Figure 7.1.
1
1
Current A-to-D and D-to-A converters use oversampling techniques to implement the
functions depicted in Figure 7.1.
95
96 Digital Speech Coding
Sampler
Quantizer
Q{ }
) (t x
c
] [n x ] [ ˆ n x
Encoder
] [n c
A-to-D Converter
Decoder
] [n c′
] [ ˆ n x′


T
Reconstruction
Filter
D-to-A Converter
) ( ˆ t x′
c
Fig. 7.1 The operations of sampling and quantization.
The sampler produces a sequence of numbers x[n] = x
c
(nT), where
T is the sampling period and f
s
= 1/T is the sampling frequency. The-
oretically, the samples x[n] are real numbers that are useful for theo-
retical analysis, but never available for computations. The well known
Shannon sampling theorem states that a bandlimited signal can be
reconstructed exactly from samples taken at twice the highest frequency
in the input signal spectrum (typically between 7 and 20 kHz for speech
and audio signals) [89]. Often, we use a lowpass filter to remove spectral
information above some frequency of interest (e.g., 4 kHz) and then use
a sampling rate such as 8000 samples/s to avoid aliasing distortion [89].
Figure 7.2 depicts a typical quantizer definition. The quantizer sim-
ply takes the real number inputs x[n] and assigns an output ˆ x[n] accord-
ing to the nonlinear discrete-output mapping Q¦ ¦.
2
In the case of
the example of Figure 7.2, the output samples are mapped to one
of eight possible values, with samples within the peak-to-peak range
being rounded and samples outside the range being “clipped” to either
the maximum positive or negative level. For samples within range, the
quantization error, defined as
e[n] = ˆ x[n] − x[n] (7.1)
2
Note that in this chapter, the notation ˆ x[n] denotes the quantized version of x[n], not the
complex cepstrum of x[n] as in Chapter 5.
7.1 Sampling and Quantization of Speech (PCM) 97
Fig. 7.2 8-level mid-tread quantizer.
satisfies the condition
−∆/2 < e[n] ≤ ∆/2, (7.2)
where ∆ is the quantizer step size. A B-bit quantizer such as the one
shown in Figure 7.2 has 2
B
levels (1 bit usually signals the sign).
Therefore, if the peak-to-peak range is 2X
m
, the step size will be
∆ = 2X
m
/2
B
. Because the quantization levels are uniformly spaced by
∆, such quantizers are called uniform quantizers. Traditionally, repre-
sentation of a speech signal by binary-coded quantized samples is called
pulse-code modulation (or just PCM) because binary numbers can be
represented for transmission as on/off pulse amplitude modulation.
The block marked “encoder” in Figure 7.1 represents the assigning
of a binary code word to each quantization level. These code words rep-
resent the quantized signal amplitudes, and generally, as in Figure 7.2,
these code words are chosen to correspond to some convenient binary
number system such that arithmetic can be done on the code words as
98 Digital Speech Coding
if they were proportional to the signal samples.
3
In cases where accu-
rate amplitude calibration is desired, the step size could be applied as
depicted in the block labeled “Decoder” at the bottom of Figure 7.1.
4
Generally, we do not need to be concerned with the fine distinction
between the quantized samples and the coded samples when we have
a fixed step size, but this is not the case if the step size is adapted
from sample-to-sample. Note that the combination of the decoder
and lowpass reconstruction filter represents the operation of D-to-A
conversion [89].
The operation of quantization confronts us with a dilemma. Most
signals, speech especially, have a wide dynamic range, i.e., their ampli-
tudes vary greatly between voiced and unvoiced sounds and between
speakers. This means that we need a large peak-to-peak range so as to
avoid clipping of the loudest sounds. However, for a given number of
levels (bits), the step size ∆ = 2X
m
/2
B
is proportional to the peak-to-
peak range. Therefore, as the peak-to-peak range increases, (7.2) states
that the size of the maximum quantization error grows. Furthermore,
for a uniform quantizer, the maximum size of the quantization error
is the same whether the signal sample is large or small. For a given
peak-to-peak range for a quantizer such as the one in Figure 7.2 we
can decrease the quantization error only by adding more levels (bits).
This is the fundamental problem of quantization.
The data rate (measured in bits/second) of sampled and quantized
speech signals is I = B f
s
. The standard values for sampling and quan-
tizing sound signals (speech, singing, instrumental music) are B = 16
and f
s
= 44.1 or 48 kHz. This leads to a bitrate of I = 16 44100 =
705, 600 bits/s.
5
This value is more than adequate and much more than
desired for most speech communication applications. The bit rate can
be lowered by using fewer bits/sample and/or using a lower sampling
rate; however, both these simple approaches degrade the perceived
quality of the speech signal. This chapter deals with a wide range of
3
In a real A-to-D converter, the sampler, quantizer, and coder are all integrated into one
system, and only the binary code words c[n] are available.
4
The

denotes the possibility of errors in the codewords. Symbol errors would cause addi-
tional error in the reconstructed samples.
5
Even higher rates (24 bits and 96 kHz sampling rate) are used for high-quality audio.
7.1 Sampling and Quantization of Speech (PCM) 99
techniques for significantly reducing the bit rate while maintaining an
adequate level of speech quality.
7.1.1 Uniform Quantization Noise Analysis
A more quantitative description of the effect of quantization can be
obtained using random signal analysis applied to the quantization error.
If the number of bits in the quantizer is reasonably high and no clip-
ping occurs, the quantization error sequence, although it is completely
determined by the signal amplitudes, nevertheless behaves as if it is a
random signal with the following properties [12]:
(Q.1) The noise samples appear to be
6
uncorrelated with the
signal samples.
(Q.2) Under certain assumptions, notably smooth input prob-
ability density functions and high rate quantizers, the
noise samples appear to be uncorrelated from sample-to-
sample, i.e., e[n] acts like a white noise sequence.
(Q.3) The amplitudes of noise samples are uniformly dis-
tributed across the range −∆/2 < e[n] ≤ ∆/2, resulting
in average power σ
2
e
= ∆
2
/12.
These simplifying assumptions allow a linear analysis that yields accu-
rate results if the signal is not too coarsely quantized. Under these con-
ditions, it can be shown that if the output levels of the quantizer are
optimized, then the quantizer error will be uncorrelated with the quan-
tizer output (however not the quantizer input, as commonly stated).
These results can be easily shown to hold in the simple case of a uni-
formly distributed memoryless input and Bennett has shown how the
result can be extended to inputs with smooth densities if the bit rate
is assumed high [12].
With these assumptions, it is possible to derive the following
formula for the signal-to-quantizing-noise ratio (in dB) of a B-bit
6
By “appear to be” we mean that measured correlations are small. Bennett has shown that
the correlations are only small because the error is small and, under suitable conditions,
the correlations are equal to the negative of the error variance [12]. The condition “uncor-
related” often implies independence, but in this case the error is a deterministic function
of the input and hence it cannot be independent of the input.
100 Digital Speech Coding
uniform quantizer:
SNR
Q
= 10log

σ
2
x
σ
2
e

= 6.02B + 4.78 − 20log
10

X
m
σ
x

, (7.3)
where σ
x
and σ
e
are the rms values of the input signal and quantization
noise samples, respectively. The formula of Equation (7.3) is an increas-
ingly good approximation as the bit rate (or the number of quantization
levels) gets large. It can be way off, however, if the bit rate is not large
[12]. Figure 7.3 shows a comparison of (7.3) with signal-to-quantization-
noise ratios measured for speech signals. The measurements were done
by quantizing 16-bit samples to 8, 9, and 10 bits. The faint dashed
lines are from (7.3) and the dark dashed lines are measured values for
uniform quantization. There is good agreement between these graphs
indicating that (7.3) is a reasonable estimate of SNR.
Note that X
m
is a fixed parameter of the quantizer, while σ
x
depends on the input signal level. As signal level increases, the ratio
X
m

x
decreases moving to the left in Figure 7.3. When σ
x
gets close
to X
m
, many samples are clipped, and the assumptions underlying
(7.3) no longer hold. This accounts for the precipitous fall in SNR for
1 < X
m

x
< 8.
0
10
20
30
40
50
60
8
9
10
Comparison of -Law and Uniform Quantizers
S
N
R
i
n
d
B
Fig. 7.3 Comparison of µ-law and linear quantization for B = 8, 9, 10: Equation (7.3) — light
dashed lines. Measured uniform quantization SNR — dark dashed lines. µ-law (µ = 100)
compression — solid lines.
7.1 Sampling and Quantization of Speech (PCM) 101
Also, it should be noted that (7.3) and Figure 7.3 show that with
all other parameters being fixed, increasing B by 1 bit (doubling the
number of quantization levels) increases the SNR by 6 dB. On the other
hand, it is also important to note that halving σ
x
decreases the SNR
by 6 dB. In other words, cutting the signal level in half is like throwing
away one bit (half of the levels) of the quantizer. Thus, it is exceed-
ingly important to keep input signal levels as high as possible without
clipping.
7.1.2 µ-Law Quantization
Also note that the SNR curves in Figure 7.3 decrease linearly with
increasing values of log
10
(X
m

x
). This is because the size of the quan-
tization noise remains the same as the signal level decreases. If the
quantization error were proportional to the signal amplitude, the SNR
would be constant regardless of the size of the signal. This could be
achieved if log(x[n]) were quantized instead of x[n], but this is not pos-
sible because the log function is ill-behaved for small values of x[n]. As
a compromise, µ-law compression (of the dynamic range) of the signal,
defined as
y[n] = X
m
log

1 + µ
|x[n]|
Xm

log(1 + µ)
sign(x[n]) (7.4)
can be used prior to uniform quantization [118]. Quantization of µ-law
compressed signals is often termed log-PCM. If µ is large (> 100) this
nonlinear transformation has the effect of distributing the effective
quantization levels uniformly for small samples and logarithmically over
the remaining range of the input samples. The flat curves in Figure 7.3
show that the signal-to-noise ratio with µ-law compression remains rel-
atively constant over a wide range of input levels.
µ-law quantization for speech is standardized in the CCITT
G.711 standard. In particular, 8-bit, µ = 255, log-PCM with f
s
=
8000 samples/s is widely used for digital telephony. This is an exam-
ple of where speech quality is deliberately compromised in order to
achieve a much lower bit rate. Once the bandwidth restriction has been
imposed, 8-bit log-PCM introduces little or no perceptible distortion.
102 Digital Speech Coding
This configuration is often referred to as “toll quality” because when
it was introduced at the beginning of the digital telephony era, it was
perceived to render speech equivalent to speech transmitted over the
best long distance lines. Nowadays, readily available hardware devices
convert analog signals directly into binary-coded µ-law samples and
also expand compressed samples back to uniform scale for conver-
sion back to analog signals. Such an A-to-D/D-to-A device is called
a encoder/decoder or “codec.” If the digital output of a codec is used
as input to a speech processing algorithm, it is generally necessary to
restore the linearity of the amplitudes through the inverse of (7.4).
7.1.3 Non-Uniform and Adaptive Quantization
µ-law compression is an example of non-uniform quantization. It is
based on the intuitive notion of constant percentage error. A more rig-
orous approach is to design a non-uniform quantizer that minimizes the
mean-squared quantization error. To do this analytically, it is necessary
to know the probability distribution of the signal sample values so that
the most probable samples, which for speech are the low amplitude
samples, will incur less error than the least probable samples. To apply
this idea to the design of a non-uniform quantizer for speech requires
an assumption of an analytical form for the probability distribution or
some algorithmic approach based on measured distributions. The fun-
damentals of optimum quantization were established by Lloyd [71] and
Max [79]. Paez and Glisson [91] gave an algorithm for designing opti-
mum quantizers for assumed Laplace and Gamma probability densities,
which are useful approximations to measured distributions for speech.
Lloyd [71] gave an algorithm for designing optimum non-uniform quan-
tizers based on sampled speech signals.
7
Optimum non-uniform quan-
tizers can improve the signal-to-noise ratio by as much as 6 dB over
µ-law quantizers with the same number of bits, but with little or no
improvement in perceived quality of reproduction, however.
7
Lloyd’s work was initially published in a Bell Laboratories Technical Note with portions
of the material having been presented at the Institute of Mathematical Statistics Meeting
in Atlantic City, New Jersey in September 1957. Subsequently this pioneering work was
published in the open literature in March 1982 [71].
7.2 Digital Speech Coding 103
Another way to deal with the wide dynamic range of speech is to let
the quantizer step size vary with time. When the quantizer in a PCM
system is adaptive, the system is called an adaptive PCM (APCM)
system. For example, the step size could satisfy
∆[n] = ∆
0
(E
ˆ n
)
1/2
, (7.5)
where ∆
0
is a constant and E
ˆ n
is the short-time energy of the speech
signal defined, for example, as in (4.6). Such adaption of the step size
is equivalent to fixed quantization of the signal after division by the
rms signal amplitude (E
ˆ n
)
1/2
. With this definition, the step size will
go up and down with the local rms amplitude of the speech signal.
The adaptation speed can be adjusted by varying the analysis win-
dow size. If the step size control is based on the unquantized signal
samples, the quantizer is called a feed-forward adaptive quantizer, and
the step size information must be transmitted or stored along with the
quantized sample, thereby adding to the information rate. To amortize
this overhead, feedforward quantization is usually applied to blocks of
speech samples. On the other hand, if the step size control is based on
past quantized samples, the quantizer is a feedback adaptive quantizer.
Jayant [57] studied a class of feedback adaptive quantizers where the
step size ∆[n] is a function of the step size for the previous sample,
∆[n − 1]. In this approach, if the previous sample is quantized using
step size ∆[n − 1] to one of the low quantization levels, then the step
size is decreased for the next sample. On the other hand, the step size
is increased if the previous sample was quantized in one of the highest
levels. By basing the adaptation on the previous sample, it is not nec-
essary to transmit the step size. It can be derived at the decoder by
the same algorithm as used for encoding.
Adaptive quantizers can achieve about 6 dB improvement (equiva-
lent to adding one bit) over fixed quantizers [57, 58].
7.2 Digital Speech Coding
As we have seen, the data rate of sampled and quantized speech is
B f
s
. Virtually perfect perceptual quality is achievable with high data
104 Digital Speech Coding
rate.
8
By reducing B or f
s
, the bit rate can be reduced, but percep-
tual quality may suffer. Reducing f
s
requires bandwidth reduction,
and reducing B too much will introduce audible distortion that may
resemble random noise. The main objective in digital speech coding
(sometimes called speech compression) is to lower the bit rate while
maintaining an adequate level of perceptual fidelity. In addition to the
two dimensions of quality and bit rate, the complexity (in terms of
digital computation) is often of equal concern.
Since the 1930s, engineers and speech scientists have worked toward
the ultimate goal of achieving more efficient representations. This work
has ledtoabasic approachinspeechcodingthat is basedonthe use of DSP
techniques and the incorporation of knowledge of speech production and
perception into the quantization process. There are many ways that this
can be done. Traditionally, speech coding methods have been classified
according to whether they attempt to preserve the waveform of the
speech signal or whether they only seek to maintain an acceptable level
of perceptual quality and intelligibility. The former are generally called
waveform coders, and the latter model-based coders. Model-based coders
are designed for obtaining efficient digital representations of speech and
only speech. Straightforward sampling and uniformquantization (PCM)
is perhaps the only pure waveform coder. Non-uniform quantization
and adaptive quantization are simple attempts to build in properties
of the speech signal (time-varying amplitude distribution) and speech
perception (quantization noise is masked by loud sounds), but these
extensions still are aimed at preserving the waveform.
Most modern model-based coding methods are based on the notion
that the speech signal can be represented in terms of an excitation sig-
nal and a time-varying vocal tract system. These two components are
quantized, and then a “quantized” speech signal can be reconstructed
by exciting the quantized filter with the quantized excitation. Figure 7.4
illustrates why linear predictive analysis can be fruitfully applied for
model-based coding. The upper plot is a segment of a speech signal.
8
If the signal that is reconstructed from the digital representation is not perceptibly different
from the original analog speech signal, the digital representation is often referred to as
being “transparent.”
7.2 Digital Speech Coding 105
0 100 200 300 400
−1
0
1
0 100 200 300 400
−0.2
0
0.2
Speech Waveform
Prediction Error Sequence
sample index
Fig. 7.4 Speech signal and corresponding prediction error signal for a 12th-order predictor.
The lower plot is the output of a linear prediction error filter A(z)
(with p = 12) that was derived from the given segment by the tech-
niques of Chapter 6. Note that the prediction error sequence amplitude
of Figure 7.4 is a factor of five lower than that of the signal itself, which
means that for a fixed number of bits, this segment of the prediction
error could be quantized with a smaller step size than the waveform
itself. If the quantized prediction error (residual) segment is used as
input to the corresponding vocal tract filter H(z) = 1/A(z), an approx-
imation to the original segment of speech would be obtained.
9
While
direct implementation of this idea does not lead to a practical method
of speech coding, it is nevertheless suggestive of more practical schemes
which we will discuss.
Today, the line between waveform coders and model-based coders
is not distinct. A more useful classification of speech coders focusses
on how the speech production and perception models are incorporated
into the quantization process. One class of systems simply attempts to
extract an excitation signal and a vocal tract system from the speech
signal without any attempt to preserve a relationship between the
waveforms of the original and the quantized speech. Such systems are
attractive because they can be implemented with modest computation.
9
To implement this time-varying inverse filtering/reconstruction requires care in fitting the
segments of the signals together at the block boundaries. Overlap-add methods [89] can
be used effectively for this purpose.
106 Digital Speech Coding
These coders, which we designate as open-loop coders, are also called
vocoders (voice coder) since they are based on the principles estab-
lished by H. Dudley early in the history of speech processing research
[30]. A second class of coders employs the source/system model for
speech production inside a feedback loop, and thus are called closed-
loop coders. These compare the quantized output to the original input
and attempt to minimize the difference between the two in some pre-
scribed sense. Differential PCM systems are simple examples of this
class, but increased availability of computational power has made it
possible to implement much more sophisticated closed-loop systems
called analysis-by-synthesis coders. Closed loop systems, since they
explicitly attempt to minimize a time-domain distortion measure, often
do a good job of preserving the speech waveform while employing many
of the same techniques used in open-loop systems.
7.3 Closed-Loop Coders
7.3.1 Predictive Coding
The essential features of predictive coding of speech were set forth in a
classic paper by Atal and Schroeder [5], although the basic principles of
predictive quantization were introduced by Cutler [26]. Figure 7.5 shows
a general block diagram of a large class of speech coding systems that
are called adaptive differential PCM (ADPCM) systems. These systems
are generally classified as waveform coders, but we prefer to emphasize
that they are closed-loop, model-based systems that also preserve the
waveform. The reader should ignore initially the blocks concerned with
adaptation and all the dotted control lines and focus instead on the core
DPCM system, which is comprised of a feedback structure that includes
the blocks labeled Q and P. The quantizer Q can have a number of
levels ranging from 2 to much higher, it can be uniform or non-uniform,
and it can be adaptive or not, but irrespective of the type of quantizer,
the quantized output can be expressed as
ˆ
d[n] = d[n] + e[n], where e[n]
is the quantization error. The block labeled P is a linear predictor, so
˜ x[n] =
p
¸
k=1
α
k
ˆ x[n − k], (7.6)
7.3 Closed-Loop Coders 107
Fig. 7.5 General block diagram for adaptive differential PCM (ADPCM).
i.e., the signal ˜ x[n] is predicted based on p past samples of the signal
ˆ x[n].
10
The signal d[n] = x[n] − ˜ x[n], the difference between the input
and the predicted signal, is the input to the quantizer. Finally, the
10
In a feed-forward adaptive predictor, the predictor parameters are estimated from the
input signal x[n], although, as shown in Figure 7.5, they can also be estimated from past
samples of the reconstructed signal ˆ x[n]. The latter would be termed a feedback adaptive
predictor.
108 Digital Speech Coding
relationship between ˆ x[n] and
ˆ
d[n] is
ˆ x[n] =
p
¸
k=1
α
k
ˆ x[n − k] +
ˆ
d[n], (7.7)
i.e., the signal ˆ x[n] is the output of what we have called the “vocal
tract filter,” and
ˆ
d[n] is the quantized excitation signal. Finally, since
ˆ x[n] = ˜ x[n] +
ˆ
d[n] it follows that
ˆ x[n] = x[n] + e[n]. (7.8)
This is the key result for DPCM systems. It says that no matter what
predictor is used in Figure 7.5, the quantization error in ˆ x[n] is iden-
tical to the quantization error in the quantized excitation signal
ˆ
d[n].
If the prediction is good, then the variance of d[n] will be less than
the variance of x[n] so it will be possible to use a smaller step size and
therefore to reconstruct ˆ x[n] with less error than if x[n] were quan-
tized directly. The feedback structure of Figure 7.5 contains within it a
source/system speech model, which, as shown in Figure 7.5(b), is the
system needed to reconstruct the quantized speech signal ˆ x[n] from the
coded difference signal input.
From (7.8), the signal-to-noise ratio of the ADPCM system is
SNR = σ
2
x

2
e
. A simple, but informative modification leads to
SNR =
σ
2
x
σ
2
d

σ
2
d
σ
2
e
= G
P
SNR
Q
, (7.9)
where the obvious definitions apply.
SNR
Q
is the signal-to-noise ratio of the quantizer, which, for fine
quantization (i.e., large number of quantization levels with small step
size), is given by (7.3). The number of bits B determines the bit-rate
of the coded difference signal. As shown in Figure 7.5, the quantizer
can be either feedforward- or feedback-adaptive. As indicated, if the
quantizer is feedforward adaptive, step size information must be part
of the digital representation.
The quantity G
P
is called the prediction gain, i.e., if G
P
> 1 it
represents an improvement gained by placing a predictor based on the
speech model inside the feedback loop around the quantizer. Neglecting
7.3 Closed-Loop Coders 109
the correlation between the signal and the quantization error, it can be
shown that
G
P
=
σ
2
x
σ
2
d
=
1
1 −
p
¸
k=1
α
k
ρ[k]
, (7.10)
where ρ[k] = φ[k]/φ[0] is the normalized autocorrelation function used
to compute the optimum predictor coefficients. The predictor can be
either fixed or adaptive, and adaptive predictors can be either feed-
forward or feedback (derived by analyzing past samples of the quan-
tized speech signal reconstructed as part of the coder). Fixed predictors
can be designed based on long-term average correlation functions. The
lower figure in Figure 7.6 shows an estimated long-term autocorrela-
tion function for speech sampled at an 8 kHz sampling rate. The upper
plot shows 10log
10
G
P
computed from (7.10) as a function of predictor
order p. Note that a first-order fixed predictor can yield about 6 dB
prediction gain so that either the quantizer can have 1 bit less or the
0 2 4 6 8 10 12
0
5
10
G
p

i
n

d
B
0 2 4 6 8 10 12
−0.5
0
0.5
1
predictor order p
autocorrelation lag
Fig. 7.6 Long-term autocorrelation function ρ[m] (lower plot) and corresponding prediction
gain G
P
.
110 Digital Speech Coding
reconstructed signal can have a 6 dB higher SNR. Higher-order fixed
predictors can achieve about 4 dB more prediction gain, and adapting
the predictor at a phoneme rate can produce an additional 4 dB of gain
[84]. Great flexibility is inherent in the block diagram of Figure 7.5, and
not surprisingly, many systems based upon the basic principle of dif-
ferential quantization have been proposed, studied, and implemented
as standard systems. Here we can only mention a few of the most
important.
7.3.2 Delta Modulation
Delta modulation (DM) systems are the simplest differential coding
systems since they use a 1-bit quantizer and usually only a first-order
predictor. DM systems originated in the classic work of Cutler [26]
and deJager [28]. While DM systems evolved somewhat independently
of the more general differential coding methods, they nevertheless fit
neatly into the general theory of predictive coding. To see how such
systems can work, note that the optimum predictor coefficient for a
first-order predictor is α
1
= ρ[1], so from (7.10) it follows that the pre-
diction gain for a first-order predictor is
G
P
=
1
1 − ρ
2
[1]
. (7.11)
Furthermore, note the nature of the long-term autocorrelation function
in Figure 7.6. This correlation function is for f
s
= 8000 samples/s. If
the same bandwidth speech signal were over-sampled at a sampling
rate higher than 8000 samples/s, we could expect that the correlation
value ρ[1] would lie on a smooth curve interpolating between samples
0 and 1 in Figure 7.6, i.e., as f
s
gets large, both ρ[1] and α
1
approach
unity and G
P
gets large. Thus, even though a 1-bit quantizer has a
very low SNR, this can be compensated by the prediction gain due to
oversampling.
Delta modulation systems can be implemented with very simple
hardware, and in contrast to more general predictive coding systems,
they are not generally implemented with block processing. Instead,
adaptation algorithms are usually based on the bit stream at the output
of the 1-bit quantizer. The simplest delta modulators use a very high
7.3 Closed-Loop Coders 111
sampling rate with a fixed predictor. Their bit rate, being equal to f
s
,
must be very high to achieve good quality reproduction. This limitation
can be mitigated with only modest increase in complexity by using an
adaptive 1-bit quantizer [47, 56]. For example, Jayant [56] showed that
for bandlimited (3 kHz) speech, adaptive delta modulation (ADM) with
bit rate (sampling rate) 40 kbits/s has about the same SNR as 6-bit
log PCM sampled at 6.6 kHz. Furthermore, he showed that doubling
the sampling rate of his ADM system improved the SNR by about
10 dB.
Adaptive delta modulation has the following advantages: it is very
simple to implement, only bit synchronization is required in transmis-
sion, it has very low coding delay, and it can be adjusted to be robust to
channel errors. For these reasons, adaptive delta modulation (ADM) is
still used for terminal-cost-sensitive digital transmission applications.
7.3.3 Adaptive Differential PCM Systems
Differential quantization with a multi-bit quantizer is called differen-
tial PCM (DPCM). As depicted in Figure 7.5, adaptive differential
PCM systems can have any combination of adaptive or fixed quantiz-
ers and/or predictors. Generally, the operations depicted in Figure 7.5
are implemented on short blocks of input signal samples, which intro-
duces delay. A careful comparison of a variety of ADPCM systems by
Noll [84] gives a valuable perspective on the relative contributions of
the quantizer and predictor. Noll compared log PCM, adaptive PCM,
fixed DPCM, and three ADPCM systems all using quantizers of 2, 3,
4, and 5-bits with 8 kHz sampling rate. For all bit rates (16, 24, 32, and
40 kbps), the results were as follows
11
:
(1) Log PCM had the lowest SNR.
(2) Adapting the quantizer (APCM) improved SNR by 6 dB.
(3) Adding first-order fixed or adaptive prediction improved the
SNR by about 4 dB over APCM.
11
The bit rates mentioned do not include overhead information for the quantized prediction
coefficients and quantizer step sizes. See Section 7.3.3.1 for a discussion of this issue.
112 Digital Speech Coding
(4) A fourth-order adaptive predictor added about 4 dB, and
increasing the predictor order to 12 added only 2 dB more.
The superior performance of ADPCM relative to log-PCM and its
relatively low computational demands have led to several standard ver-
sions (ITU G.721, G.726, G.727), which are often operated at 32 kbps
where quality is superior to log-PCM (ITU G.711) at 64 kbps.
The classic paper by Atal and Schroeder [5] contained a number
of ideas that have since been applied with great success in adaptive
predictive coding of speech. One of these concerns the quantization
noise introduced by ADPCM, which we have seen is simply added to
the input signal in the reconstruction process. If we invoke the white
noise approximation for quantization error, it is clear that the noise will
be most prominent in the speech at high frequencies where speech spec-
trum amplitudes are low. A simple solution to this problem is to pre-
emphasize the speech signal before coding, i.e., filter the speech with a
linear filter that boosts the high-frequency (HF) part of the spectrum.
After reconstruction of the pre-emphasized speech, it can be filtered
with the inverse system to restore the spectral balance, and in the
process, the noise spectrum will take on the shape of the de-emphasis
filter spectrum. A simple pre-emphasis system of this type has system
function (1 − γz
−1
) where γ is less than one.
12
A more sophisticated
application of this idea is to replace the simple fixed pre-emphasis fil-
ter with a time-varying filter designed to shape the quantization noise
spectrum. An equivalent approach is to define a new feedback cod-
ing system where the quantization noise is computed as the difference
between the input and output of the quantizer and then shaped before
adding to the prediction residual. All these approaches raise the noise
spectrum at low frequencies and lower it at high frequencies, thus tak-
ing advantage of the masking effects of the prominent low frequencies
in speech [2].
Another idea that has been applied effectively in ADPCM systems
as well as analysis-by-synthesis systems is to include a long-delay pre-
dictor to capture the periodicity as well as the short-delay correlation
12
Values around γ = 0.4 for fs = 8 kHz work best. If γ is too close to one, the low frequency
noise is emphasized too much.
7.3 Closed-Loop Coders 113
inherent in voiced speech. Including the simplest long-delay predictor
would change (7.6) to
˜ x[n] = βˆ x[n − M] +
p
¸
k=1
α
k
(ˆ x[n] − βˆ x[n − M]). (7.12)
The parameter M is essentially the pitch period (in samples) of voiced
speech, and the parameter β accounts for amplitude variations between
periods. The predictor parameters in (7.12) cannot be jointly opti-
mized. One sub-optimal approach is to estimate β and M first by
determining M that maximizes the correlation around the expected
pitch period and setting β equal to the normalized correlation at M.
Then the short-delay predictor parameters α
k
are estimated from the
output of the long-delay prediction error filter.
13
When this type of
predictor is used, the “vocal tract system” used in the decoder would
have system function
H(z) =

¸
¸
¸
¸
¸
1
1 −
p
¸
k=1
α
k
z
−k
¸

1
1 − βz
−M

. (7.13)
Adding long-delay prediction takes more information out of the speech
signal and encodes it in the predictor. Even more complicated predic-
tors with multiple delays and gains have been found to improve the
performance of ADPCM systems significantly [2].
7.3.3.1 Coding the ADPCM Parameters
A glance at Figure 7.5 shows that if the quantizer and predictor are
fixed or feedback-adaptive, then only the coded difference signal is
needed to reconstruct the speech signal (assuming that the decoder
has the same coefficients and feedback-adaptation algorithms). Other-
wise, the predictor coefficients and quantizer step size must be included
as auxiliary data (side information) along with the quantized difference
13
Normally, this analysis would be performed on the input speech signal with the resulting
predictor coefficients applied to the reconstructed signal as in (7.12).
114 Digital Speech Coding
signal. The difference signal will have the same sampling rate as the
input signal. The step size and predictor parameters will be estimated
and changed at a much lower sampling rate, e.g., 50–100 times/s. The
total bit rate will be the sum of all the bit rates.
Predictor Quantization The predictor parameters must be quan-
tized for efficient digital representation of the speech signal. As dis-
cussed in Chapter 6, the short-delay predictors can be represented in
many equivalent ways, almost all of which are preferable to direct quan-
tization of the predictor coefficients themselves. The PARCOR coeffi-
cients can be quantized with a non-uniform quantizer or transformed
with an inverse sine or hyperbolic tangent function to flatten their sta-
tistical distribution and then quantized with a fixed uniform quantizer.
Each coefficient can be allocated a number of bits contingent on its
importance in accurately representing the speech spectrum [131].
Atal [2] reported a system which used a 20th-order predictor and
quantized the resulting set of PARCOR coefficients after transforma-
tion with an inverse sine function. The number of bits per coefficient
ranged from 5 for each of the lowest two PARCORs down to 1 each for
the six highest-order PARCORs, yielding a total of 40 bits per frame.
If the parameters are up-dated 100 times/s, then a total of 4000 bps for
the short-delay predictor information is required. To save bit rate, it
is possible to up-date the short-delay predictor 50 times/s for a total
bit rate contribution of 2000 bps [2]. The long delay predictor has a
delay parameter M, which requires 7 bits to cover the range of pitch
periods to be expected with 8 kHz sampling rate of the input. The
long-delay predictor in the system mentioned above used delays of
M − 1, M, M + 1 and three gains each quantized to 4 or 5 bit accu-
racy. This gave a total of 20 bits/frame and added 2000 bps to the
overall bit rate.
Vector Quantization for Predictor Quantization Another
approach to quantizing the predictor parameters is called vector quan-
tization, or VQ [45]. The basic principle of vector quantization is
depicted in Figure 7.7. VQ can be applied in any context where a set
of parameters naturally groups together into a vector (in the sense of
7.3 Closed-Loop Coders 115
Distortion
Measure
Code Book
2
B
Vectors
Table
Lookup
Code Book
2
B
Vectors
v
i
i
k

k

i

Encoder Decoder
Fig. 7.7 Vector quantization.
linear algebra).
14
For example the predictor coefficients can be repre-
sented by vectors of predictor coefficients, PARCOR coefficients, cep-
strum values, log area ratios, or line spectrum frequencies. In VQ, a
vector, v, to be quantized is compared exhaustively to a “codebook”
populated with representative vectors of that type. The index i of the
closest vector ˆ v
i
to v, according to a prescribed distortion measure,
is returned from the exhaustive search. This index i then represents
the quantized vector in the sense that if the codebook is known, the
corresponding quantized vector ˆ v
i
can be looked up. If the codebook
has 2
B
entries, the index i can be represented by a B bit number.
The design of a vector quantizer requires a training set consisting
of a large number (L) of examples of vectors that are drawn from the
same distribution as vectors to be quantized later, along with a dis-
tortion measure for comparing two vectors. Using an iterative process,
2
B
codebook vectors are formed from the training set of vectors. The
“training phase” is computationally intense, but need only be done
once, and it is facilitated by the LBG algorithm [70]. One advantage of
VQ is that the distortion measure can be designed to sort the vectors
into perceptually relevant prototype vectors. The resulting codebook is
then used to quantize general test vectors. Typically a VQ training set
consists of at least 10 times the ultimate number of codebook vectors,
i.e., L ≥ 10 2
B
.
14
Independent quantization of individual speech samples or individual predictor coefficients
is called scalar quantization.
116 Digital Speech Coding
The quantization process is also time consuming because a given
vector must be systematically compared with all the codebook vectors
to determine which one it is closest to. This has led to numerous inno-
vations in codebook structure that speed up the look up process [45].
Difference Signal Quantization It is necessary to code the differ-
ence signal in ADPCM. Since the difference signal has the same sam-
pling rate as the input signal, it is desirable to use as few bits as possible
for the quantizer. If straightforward B-bit uniform quantization is used,
the additional bit rate would be B f
s
. This was the approach used in
earlier studies by Noll [84] as mentioned in Section 7.3.3. In order to
reduce the bit rate for coding the difference samples, some sort of block
coding must be applied. One approach is to design a multi-bit quantizer
that only operates on the largest samples of the difference signal. Atal
and Schroeder [2, 7] proposed to precede the quantizer by a “center
clipper,” which is a system that sets to zero value all samples whose
amplitudes are within a threshold band and passes those samples above
threshold unchanged. By adjusting the threshold, the number of zero
samples can be controlled. For high thresholds, long runs of zero sam-
ples result, and the entropy of the quantized difference signal becomes
less than one. The blocks of zero samples can be encoded efficiently
using variable-length-to-block codes yielding average bits/sample on
the order of 0.7 [2]. For a sampling rate of 8 kHz, this coding method
needs a total of 5600 bps for the difference signal. When this bit rate
of the differential coder is added to the bit rate for quantizing the
predictor information (≈4000 bps) the total comes to approximately
10,000 bps.
7.3.3.2 Quality vs. Bit Rate for ADPCM Coders
ADPCM systems normally operate with an 8 kHz sampling rate, which
is a compromise aimed at providing adequate intelligibility and mod-
erate bit rate for voice communication. With this sampling rate, the
speech signal must be bandlimited to somewhat less than 4 kHz by a
lowpass filter. Adaptive prediction and quantization can easily lower
the bit rate to 32 kbps with no degradation in perceived quality (with
7.3 Closed-Loop Coders 117
64 kbps log PCM toll quality as a reference). Of course, raising the
bit rate for ADPCM above 64 kbps cannot improve the quality over
log-PCM because of the inherent frequency limitation. However, the
bit rate can be lowered below 32 kbps with only modest distortion
until about 10 kbps. Below this value, quality degrades significantly.
In order to achieve near-toll-quality at low bit rates, all the techniques
discussed above and more must be brought into play. This increases
the system complexity, although not unreasonably so for current DSP
hardware. Thus, ADPCM is an attractive coding method when toll
quality is required at modest cost, and where adequate transmission
and/or storage capacity is available to support bit rates on the order
of 10 kbps.
7.3.4 Analysis-by-Synthesis Coding
While ADPCM coding can produce excellent results at moderate bit
rates, its performance is fundamentally constrained by the fact that the
difference signal has the same sampling rate as the input signal. The
center clipping quantizer produces a difference signal that can be coded
efficiently. However, this approach is clearly not optimal since the center
clipper throws away information in order to obtain a sparse sequence.
What is needed is a way of creating an excitation signal for the vocal
tract filter that is both efficient to code and also produces decoded
speech of high quality. This can be done within the same closed-loop
framework as ADPCM, but with some significant modifications.
7.3.4.1 Basic Analysis-by-Synthesis Coding System
Figure 7.8 shows the block diagram of another class of closed-loop
digital speech coders. These systems are called “analysis-by-synthesis
coders” because the excitation is built up using an iterative process to
produce a “synthetic” vocal tract filter output ˆ x[n] that matches the
input speech signal according to a perceptually weighted error crite-
rion. As in the case of the most sophisticated ADPCM systems, the
operations of Figure 7.8 are carried out on blocks of speech samples. In
particular, the difference, d[n], between the input, x[n], and the output
of the vocal tract filter, ˆ x[n], is filtered with a linear filter called the
118 Digital Speech Coding
Perceptual
Weighting
W(z)
Excitation
Generator
Vocal Tract
Filter
H(z)
+
] [n x ] [n d
] [
ˆ
n d
] [ ˆ n x
-
+
] [n d′
Fig. 7.8 Structure of analysis-by-synthesis speech coders.
perceptual weighting filter, W(z). As the first step in coding a block of
speech samples, both the vocal tract filter and the perceptual weighting
filter are derived from a linear predictive analysis of the block. Then,
the excitation signal is determined from the perceptually weighted dif-
ference signal d

[n] by an algorithm represented by the block labeled
“Excitation Generator.”
Note the similarity of Figure 7.8 to the core ADPCM diagram of
Figure 7.5. The perceptual weighting and excitation generator inside
the dotted box play the role played by the quantizer in ADPCM, where
an adaptive quantization algorithm operates on d[n] to produce a quan-
tized difference signal
ˆ
d[n], which is the input to the vocal tract system.
In ADPCM, the vocal tract model is in the same position in the closed-
loop system, but instead of the synthetic output ˆ x[n], a signal ˜ x[n]
predicted from ˆ x[n] is subtracted from the input to form the difference
signal. This is a key difference. In ADPCM, the synthetic output differs
from the input x[n] by the quantization error. In analysis-by-synthesis,
ˆ x[n] = x[n] − d[n], i.e., the reconstruction error is −d[n], and a percep-
tually weighted version of that error is minimized in the mean-squared
sense by the selection of the excitation
ˆ
d[n].
7.3.4.2 Perceptual Weighting of the Difference Signal
Since −d[n] is the error in the reconstructed signal, it is desirable to
shape its spectrum to take advantage of perceptual masking effects. In
ADPCM, this is accomplished by quantization noise feedback or preem-
phasis/deemphasis filtering. In analysis-by-synthesis coding, the input
7.3 Closed-Loop Coders 119
to the vocal tract filter is determined so as to minimize the perceptually-
weighted error d

[n]. The weighting is implemented by linear filtering,
i.e., d

[n] = d[n] ∗ w[n] with the weighting filter usually defined in terms
of the vocal tract filter as the linear system with system function
W(z) =
A(z/α
1
)
A(z/α
2
)
=
H(z/α
2
)
H(z/α
1
)
. (7.14)
The poles of W(z) lie at the same angle but at α
2
times the radii of
the poles of H(z), and the zeros of W(z) are at the same angles but at
radii of α
1
times the radii of the poles of H(z). If α
1
> α
2
the frequency
response is like a “controlled” inverse filter for H(z), which is the shape
desired. Figure 7.9 shows the frequency response of such a filter, where
typical values of α
1
= 0.9 and α
2
= 0.4 are used in (7.14). Clearly, this
filter tends to emphasize the high frequencies (where the vocal tract
filter gain is low) and it deemphasizes the low frequencies in the error
signal. Thus, it follows that the error will be distributed in frequency so
that relatively more error occurs at low frequencies, where, in this case,
such errors would be masked by the high amplitude low frequencies.
By varying α
1
and α
2
in (7.14) the relative distribution of error can be
adjusted.
0 1000 2000 3000 4000
−30
−20
−10
0
10
20
30
frequency in Hz,
l
o
g
m
a
g
n
i
t
u
d
e
i
n
d
B
Fig. 7.9 Comparison of frequency responses of vocal tract filter and perceptual weighting
filter in analysis-by-synthesis coding.
120 Digital Speech Coding
7.3.4.3 Generating the Excitation Signal
Most analysis-by-synthesis systems generate the excitation from a finite
fixed collection of input components, which we designate here as f
γ
[n]
for 0 ≤ n ≤ L − 1, where L is the excitation frame length and γ ranges
over the finite set of components. The input is composed of a finite sum
of scaled components selected from the given collection, i.e.,
ˆ
d[n] =
N
¸
k=1
β
k
f
γ
k
[n]. (7.15)
The β
k
s and the sequences f
γ
k
[n] are chosen to minimize
E =
L−1
¸
n=0
((x[n] − h[n] ∗
ˆ
d[n]) ∗ w[n])
2
, (7.16)
where h[n] is the vocal tract impulse response and w[n] is the impulse
response of the perceptual weighting filter with system function (7.14).
Since the component sequences are assumed to be known at both the
coder and decoder, the βs and γs are all that is needed to represent the
input
ˆ
d[n]. It is very difficult to solve simultaneously for the optimumβs
and γs that minimize (7.16). However, satisfactory results are obtained
by solving for the component signals one at a time.
Starting with the assumption that the excitation signal is zero dur-
ing the current excitation
15
analysis frame (indexed 0 ≤ n ≤ L − 1), the
output in the current frame due to the excitation determined for previ-
ous frames is computed and denoted ˆ x
0
[n] for 0 ≤ n ≤ L − 1. Normally
this would be a decaying signal that could be truncated after L samples.
The error signal in the current frame at this initial stage of the iterative
process would be d
0
[n] = x[n] − ˆ x
0
[n], and the perceptually weighted
error would be d

0
[n] = d
0
[n] ∗ w[n]. Now at the first iteration stage,
assume that we have determined which of the collection of input com-
ponents f
γ
1
[n] will reduce the weighted mean-squared error the most
and we have also determined the required gain β
1
. By superposition,
ˆ x
1
[n] = ˆ x
0
[n] + β
1
f
γ
1
[n] ∗ h[n]. The weighted error at the first itera-
tion is d

1
[n] = (d
0
[n] − β
1
f
γ
1
[n] ∗ h[n] ∗ w[n]), or if we define the per-
ceptually weighted vocal tract impulse response as h

[n] = h[n] ∗ w[n]
15
Several analysis frames are usually included in one linear predictive analysis frame.
7.3 Closed-Loop Coders 121
and invoke the distributive property of convolution, then the weighted
error sequence is d

1
[n] = d

0
[n] − β
1
f
γ
1
[n] ∗ h

[n]. This process is contin-
ued by finding the next component, subtracting it from the previously
computed residual error, and so forth. Generalizing to the kth iteration,
the corresponding equation takes the form:
d

k
[n] = d

k−1
[n] − β
k
y

k
[n], (7.17)
where y

[n] = f
γ
k
[n] ∗ h

[n] is the output of the perceptually weighted
vocal tract impulse response due to the input component f
γ
k
[n]. The
mean-squared error at stage k of the iteration is defined as
E
k
=
L−1
¸
n=0
(d

k
[n])
2
=
L−1
¸
n=0
(d

k−1
[n] − β
k
y

k
[n])
2
, (7.18)
where it is assumed that d

k−1
[n] is the weighted difference signal that
remains after k − 1 steps of the process.
Assuming that γ
k
is known, the value of β
k
that minimizes E
k
in
(7.18) is
β
k
=
L−1
¸
n=0
d

k−1
[n]y

k
[n]
L−1
¸
n=0
(y

[n])
2
, (7.19)
and the corresponding minimum mean-squared error is
E
(min)
k
=
L−1
¸
n=0
(d

k−1
[n])
2
− β
2
k
L−1
¸
n=0
(y

k
[n])
2
. (7.20)
While we have assumed in the above discussion that f
γ
k
[n] and
β
k
are known, we have not discussed how they can be found. Equa-
tion (7.20) suggests that the mean-squared error can be minimized by
maximizing
β
2
k
L−1
¸
n=0
(y

k
[n])
2
=

L−1
¸
n=0
d

k−1
[n]y

k
[n]

2
L−1
¸
n=0
(y

k
[n])
2
, (7.21)
122 Digital Speech Coding
which is essentially the normalized cross-correlation between the new
input component and the weighted residual error d

k−1
[n]. An exhaus-
tive search through the collection of input components will determine
which component f
γ
k
[n] will maximize the quantity in (7.21) and thus
reduce the mean-squared error the most. Once this component is found,
the corresponding β
k
can be found from (7.19), and the new error
sequence computed from (7.17).
After N iterations of this process, the complete set of component
input sequences and corresponding coefficients will have been deter-
mined.
16
If desired, the set of input sequences so determined can be
assumed and a new set of βs can be found jointly by solving a set of
N linear equations.
7.3.4.4 Multi-Pulse Excitation Linear Prediction (MPLP)
The procedure described in Section 7.3.4.3 is quite general since the
only constraint was that the collection of component input sequences be
finite. The first analysis-by-synthesis coder was called multi-pulse linear
predictive coding [4]. In this system, the component input sequences are
simply isolated unit impulse sequences, i.e., f
γ
[n] = δ[n − γ], where γ
is an integer such that 0 ≤ γ ≤ L − 1. The excitation sequence derived
by the process described in Section 7.3.4.3 is therefore of the form:
ˆ
d[n] =
N
¸
k=1
β
k
δ[n − γ
k
], (7.22)
where N is usually on the order of 4–5 impulses in a 5 ms (40 samples
for f
s
= 8 kHz) excitation analysis frame.
17
In this case, an exhaustive
search at each stage can be achieved by computing just one cross-
correlation function and locating the impulse at the maximum of the
cross-correlation. This is because the components y

k
[n] are all just
shifted versions of h

[n].
16
Alternatively, N need not be fixed in advance. The error reduction process can be stopped
when E
(min)
k
falls below a prescribed threshold.
17
It is advantageous to include two or more excitation analysis frames in one linear predic-
tive analysis frame, which normally is of 20 ms (160 samples) duration.
7.3 Closed-Loop Coders 123
In a speech coder based on multi-pulse analysis-by-synthesis, the
speech signal is represented by quantized versions of the prediction
coefficients, the pulse locations, and the pulse amplitudes. The predic-
tor coefficients can be coded as discussed for ADPCM by encoding any
of the alternative representations either by scalar or vector quantiza-
tion. A number of different approaches have been devised for coding
the excitation parameters. These usually involve some sort of differ-
ential coding scheme. At bit rates on the order of 10–16 kbps multi-
pulse coding can approach toll quality; however, below about 10 kbps,
quality degrades rapidly due to the many parameters that must be
coded [13].
Regular-pulse excitation (RPE) is a special case of multipulse exci-
tation where it is easier to encode the impulse locations. Specifically
after determining the location γ
1
of the first impulse, all other impulses
are located at integer multiples of a fixed spacing on either side of γ
1
.
Thus, only γ
1
, the integer multiples, and the gain constants must be
encoded. Using a pre-set spacing of 4 and an excitation analysis win-
dow of 40 samples yields high quality at about 10 kbps [67]. Because the
impulse locations are fixed after the first iteration, RPE requires signifi-
cantly less computation than full-blown multipulse excitation analysis.
One version of RPE is the basis for the 13 kbps digital coder in the
GSM mobile communications system.
7.3.4.5 Code-Excited Linear Prediction (CELP)
Since the major portion of the bit rate of an analysis-by-synthesis coder
lies in the excitation signal, it is not surprising that a great deal of
effort has gone into schemes for finding excitation signals that are eas-
ier to encode than multipulse excitation, but yet maintain high qual-
ity of reproduction of the speech signal. The next major innovation
in the history of analysis-by-synthesis systems was called code-excited
linear predictive coding or CELP [112].
18
In this scheme, the excita-
tion signal components are Gaussian random sequences stored in a
18
Earlier work by Stewart [123] used a residual codebook (populated by the Lloyd algorithm
[71]) with a low complexity trellis search.
124 Digital Speech Coding
“codebook.”
19
If the codebook contains 2
M
sequences, then a given
sequence can be specified by an M-bit number. Typical codebook sizes
are 256, 512 or 1024. Within the codebook, the sequence lengths are
typically L = 40–60 samples (5–7.5 ms at 8 kHz sampling rate). If N
sequences are used to form the excitation, then a total of M N bits
will be required to code the sequences. Additional bits are still required
to code the β
k
s. With this method, the decoder must also have a copy
of the analysis codebook so that the excitation can be regenerated at
the decoder.
The main disadvantage of CELP is the high computational cost of
exhaustively searching the codebook at each stage of the error min-
imization. This is because each codebook sequence must be filtered
with the perceptually weighted impulse response before computing the
cross correlation with the residual error sequence. Efficient searching
schemes and structured codebooks have eased the computational bur-
den, and modern DSP hardware can easily implement the computations
in real time.
The basic CELP framework has been applied in the development of
numerous speech coders that operate with bit rates in the range from
4800 to 16000 bps. The Federal Standard 1016 (FS1016) was adopted
by the Department of Defense for use in secure voice transmission
at 4800 bps. Another system called VSELP (vector sum excited linear
prediction), which uses multiple codebooks to achieve a bit rate of
8000 bps, was adopted in 1989 for North American Cellular Systems.
The GSM half rate coder operating at 5600 bps is based on the IS-54
VSELP coder. Due to the block processing structure, CELP coders
can introduce more than 40 ms delay into communication systems. The
ITU G.728 low-delay CELP coder uses very short excitation analysis
frames and backward linear prediction to achieve high quality at a bit
rate of 16,000 bps with delays less than 2 ms. Still another standardized
CELP system is the ITU G.729 conjugate-structure algebraic CELP
(CS-ACELP) coder that is used in some mobile communication systems.
19
While it may seem counter-intuitive that the excitation could be comprised of white noise
sequence, remember that, in fact, the object of linear prediction is to produce just such
a signal.
7.4 Open-Loop Coders 125
7.3.4.6 Long-Delay Predictors in Analysis-by-Synthesis
Coders
Long-delay predictors have an interesting interpretation in analysis-by-
synthesis coding. If we maintain a memory of one or more frames of
previous values of the excitation signal, we can add a term β
0
ˆ
d[n − γ
0
]
to (7.15). The long segment of the past excitation acts as a sort of
codebook where the L-sample codebook sequences overlap by L − 1
samples. The first step of the excitation computation would be to com-
pute γ
0
using (7.21) and then β
0
using (7.19). For each value of γ
0
to be
tested, this requires that the weighted vocal tract impulse response be
convolved with L-sample segments of the past excitation starting with
sample γ
0
. This can be done recursively to save computation. Then the
component β
0
ˆ
d[n − γ
0
] can be subtracted from the initial error to start
the iteration process in either the multipulse or the CELP framework.
The additional bit rate for coding β
0
and γ
0
is often well worthwhile,
and long-delay predictors are used in many of the standardized coders
mentioned above. This incorporation of components of the past history
of the excitation has been referred to as “self-excitation” [104] or the
use of an “adaptive codebook” [19].
7.4 Open-Loop Coders
ADPCM coding has not been useful below about 9600 bps, multipulse
coding has similar limitations, and CELP coding has not been used at
bit rates below about 4800 bits/s. While these closed-loop systems have
many attractive features, it has not been possible to generate excitation
signals that can be coded at low bit rates and also produce high quality
synthetic speech. For bit rates below 4800 bps, engineers have turned
to the vocoder principles that were established decades ago. We call
these systems open-loop systems because they do not determine the
excitation by a feedback process.
7.4.1 The Two-State Excitation Model
Figure 7.10 shows a source/system model for speech that is closely
related to physical models for speech production. As we have discussed
126 Digital Speech Coding
Fig. 7.10 Two-state-excitation model for speech synthesis.
in the previous chapters, the excitation model can be very simple.
Unvoiced sounds are produced by exciting the system with white noise,
and voiced sounds are produced by a periodic impulse train excitation,
where the spacing between impulses is the pitch period, P
0
. We shall
refer to this as the two-state excitation model. The slowly-time-varying
linear system models the combined effects of vocal tract transmission,
radiation at the lips, and, in the case of voiced speech, the lowpass
frequency shaping of the glottal pulse. The V/UV (voiced/unvoiced
excitation) switch produces the alternating voiced and unvoiced seg-
ments of speech, and the gain parameter G controls the level of the
filter output. When values for V/UV decision, G, P
0
, and the param-
eters of the linear system are supplied at periodic intervals (frames)
then the model becomes a speech synthesizer, as discussed in Chap-
ter 8. When the parameters of the model are estimated directly
from a speech signal, the combination of estimator and synthesizer
becomes a vocoder or, as we prefer, an open-loop analysis/synthesis
speech coder.
7.4.1.1 Pitch, Gain, and V/UV Detection
The fundamental frequency of voiced speech can range from well below
100 Hz for low-pitched male speakers to over 250 Hz for high-pitched
voices of women and children. The fundamental frequency varies slowly,
with time, more or less at the same rate as the vocal tract motions. It
is common to estimate the fundamental frequency F
0
or equivalently,
7.4 Open-Loop Coders 127
the pitch period P
0
= 1/F
0
at a frame rate of about 50–100 times/s.
To do this, short segments of speech are analyzed to detect periodicity
(signaling voiced speech) or aperiodicity (signaling unvoiced speech).
One of the simplest, yet most effective approaches to pitch detection,
operates directly on the time waveform by locating corresponding peaks
and valleys in the waveform, and measuring the times between the
peaks [43]. The STACF is also useful for this purpose as illustrated
in Figure 4.8, which shows autocorrelation functions for both voiced
speech, evincing a peak at the pitch period, and unvoiced speech, which
shows no evidence of periodicity. In pitch detection applications of the
short-time autocorrelation, it is common to preprocess the speech by
a spectrum flattening operation such as center clipping [96] or inverse
filtering [77]. This preprocessing tends to enhance the peak at the pitch
period for voiced speech while suppressing the local correlation due to
formant resonances.
Another approach to pitch detection is suggested by Figure 5.5,
which shows the cepstra of segments of voiced and unvoiced speech. In
this case, a strong peak in the expected pitch period range signals voiced
speech, and the location of the peak is the pitch period. Similarly, lack
of a peak in the expected range signals unvoiced speech [83]. Figure 5.6
shows a sequence of short-time cepstra which moves from unvoiced to
voiced speech in going from the bottom to the top of the figure. The
time variation of the pitch period is evident in the upper plots.
The gain parameter G is also found by analysis of short segments
of speech. It should be chosen so that the short-time energy of the syn-
thetic output matches the short-time energy of the input speech signal.
For this purpose, the autocorrelation function value φ
ˆ n
[0] at lag 0 or
the cepstrum value c
ˆ n
[0] at quefrency 0 can be used to determine the
energy of the segment of the input signal.
For digital coding applications, the pitch period, V/UV, and gain
must be quantized. Typical values are 7 bits for pitch period (P
0
= 0
signals UV) and 5 bits for G. For a frame rate of 50 frames/s, this totals
600 bps, which is well below the bit rate used to encode the excitation
signal in closed-loop coders such as ADPCM or CELP. Since the vocal
tract filter can be coded as in ADPCM or CELP, much lower total bit
128 Digital Speech Coding
rates are common in open-loop systems. This comes at a large cost in
quality of the synthetic speech output, however.
7.4.1.2 Vocal Tract System Estimation
The vocal tract system in the synthesizer of Figure 7.10 can take many
forms. The primary methods that have been used have been homomor-
phic filtering and linear predictive analysis as discussed in Chapters 5
and 6, respectively.
The Homomorphic Vocoder Homomorphic filtering can be used
to extract a sequence of impulse responses from the sequence of cep-
stra that result from short-time cepstrum analysis. Thus, one cepstrum
computation can yield both an estimate of pitch and the vocal tract
impulse response. In the original homomorphic vocoder, the impulse
response was digitally coded by quantizing each cepstrum value indi-
vidually (scalar quantization) [86]. The impulse response, reconstructed
from the quantized cepstrum at the synthesizer, is simply convolved
with the excitation created from the quantized pitch, voicing, and gain
information, i.e., s[n] = Ge[n] ∗ h[n].
20
In a more recent application of
homomorphic analysis in an analysis-by-synthesis framework [21], the
cepstrum values were coded using vector quantization, and the excita-
tion derived by analysis-by-synthesis as described in Section 7.3.4.3. In
still another approach to digital coding, homomorphic filtering was used
to remove excitation effects in the short-time spectrum and then three
formant frequencies were estimated from the smoothed spectra. Fig-
ure 5.6 shows an example of the formants estimated for voiced speech.
These formant frequencies were used to control the resonance frequen-
cies of a synthesizer comprised of a cascade of second-order section IIR
digital filters [111]. Such a speech coder is called a formant vocoder.
LPC Vocoder Linear predictive analysis can also be used to esti-
mate the vocal tract system for an open-loop coder with two-state
20
Care must be taken at frame boundaries. For example, the impulse response can be
changed at the time a new pitch impulse occurs, and the resulting output can overlap
into the next frame.
7.4 Open-Loop Coders 129
excitation [6]. In this case, the prediction coefficients can be coded in
one of the many ways that we have already discussed. Such analy-
sis/synthesis coders are called LPC vocoders. A vocoder of this type
was standardized by the Department of Defense as the Federal Stan-
dard FS1015. This system is also called LPC-10 (or LPC-10e) because
a 10th-order covariance linear predictive analysis is used to estimate
the vocal tract system. The LPC-10 system has a bit rate of 2400 bps
using a frame size of 22.5 ms with 12 bits/frame allocated to pitch,
voicing and gain and the remaining bits allocated to the vocal tract
filter coded as PARCOR coefficients.
7.4.2 Residual-Excited Linear Predictive Coding
In Section 7.2, we presented Figure 7.4 as motivation for the use of
the source/system model in digital speech coding. This figure shows
an example of inverse filtering of the speech signal using a predic-
tion error filter, where the prediction error (or residual) is significantly
smaller and less lowpass in nature. The speech signal can be recon-
structed from the residual by passing it through the vocal tract sys-
tem H(z) = 1/A(z). None of the methods discussed so far attempt
to directly code the prediction error signal in an open-loop manner.
ADPCM and analysis-by-synthesis systems derive the excitation to the
synthesis filter by a feedback process. The two-state model attempts
to construct the excitation signal by direct analysis and measurement
of the input speech signal. Systems that attempt to code the resid-
ual signal directly are called residual-excited linear predictive (RELP)
coders.
Direct coding of the residual faces the same problem faced in
ADPCM or CELP: the sampling rate is the same as that of the input,
and accurate coding could require several bits per sample. Figure 7.11
shows a block diagram of a RELP coder [128]. In this system, which is
quite similar to voice-excited vocoders (VEV) [113, 130], the problem
of reducing the bit rate of the residual signal is attacked by reducing
its bandwidth to about 800 Hz, lowering the sampling rate, and coding
the samples with adaptive quantization. Adaptive delta modulation
was used in [128], but APCM could be used if the sampling rate is
130 Digital Speech Coding
Inverse
Filter
A(z)
Lowpass
&
Decimator
Adaptive
Quantizer
Linear
Predictive
Analysis
] [n x ] [n r
Adaptive
Quantizer
Decoder
Interpolator
& Spectrum
Flattener
Synthesis
Filter
H(z)
] [n x
] [n d
} {α
] [ ˆ n
} {α
] [n d
Encoder
Decoder
Fig. 7.11 Residual-excited linear predictive (RELP) coder and decoder.
lowered to 1600 Hz. The 800 Hz band is wide enough to contain several
harmonics of the highest pitched voices. The reduced bandwidth resid-
ual is restored to full bandwidth prior to its use as an excitation signal
by a nonlinear spectrum flattening operation, which restores higher
harmonics of voiced speech. White noise is also added according to an
empirically derived recipe. In the implementation of [128], the sam-
pling rate of the input was 6.8 kHz and the total bit rate was 9600 kbps
with 6800 bps devoted to the residual signal. The quality achieved at
this rate was not significantly better than the LPC vocoder with a
two-state excitation model. The principle advantage of this system is
that no hard V/UV decision must be made, and no pitch detection is
required. While this system did not become widely used, its basic prin-
ciples can be found in subsequent open-loop coders that have produced
much better speech quality at bit rates around 2400 bps.
7.4.3 Mixed Excitation Systems
While two-state excitation allows the bit rate to be quite low, the qual-
ity of the synthetic speech output leaves much to be desired. The output
of such systems is often described as “buzzy,” and in many cases, errors
in estimating pitch period or voicing decision cause the speech to sound
unnatural if not unintelligible. The weaknesses of the two-state model
7.4 Open-Loop Coders 131
for excitation spurred interest in a mixed excitation model where a
hard decision between V and UV is not required. Such a model was
first proposed by Makhoul et al. [75] and greatly refined by McCree
and Barnwell [80].
Figure 7.12 depicts the essential features of the mixed-excitation
linear predictive coder (MELP) proposed by McCree and Barn-
well [80]. This configuration was developed as the result of careful
experimentation, which focused one-by-one on the sources of distor-
tions manifest in the two-state excitation coder such as buzziness and
tonal distortions. The main feature is that impulse train excitation
and noise excitation are added instead of switched. Prior to their addi-
tion, they each pass through a multi-band spectral shaping filter. The
gains in each of five bands are coordinated between the two filters so
that the spectrum of e[n] is flat. This mixed excitation helps to model
short-time spectral effects such as “devoicing” of certain bands during
voiced speech.
21
In some situations, a “jitter” parameter ∆P is invoked
to better model voicing transitions. Other important features of the
MELP system are lumped into the block labeled “Enhancing Filters.”
This represents adaptive spectrum enhancement filters used to enhance
formant regions,
22
and a spectrally flat “pulse dispersion filter” whose
Fig. 7.12 Mixed-excitation linear predictive (MELP) decoder.
21
Devoicing is evident in frame 9 in Figure 5.6.
22
Such filters are also used routinely in CELP coders and are often referred to as post
filters.
132 Digital Speech Coding
purpose is to reduce “peakiness” due to the minimum-phase nature of
the linear predictive vocal tract system.
These modifications to the basic two-state excitation LPC vocoder
produce marked improvements in the quality of reproduction of the
speech signal. Several new parameters of the excitation must be esti-
mated at analysis time and coded for transmission, but these add only
slightly to either the analysis computation or the bit rate [80]. The
MELP coder is said to produce speech quality at 2400 bps that is com-
parable to CELP coding at 4800 bps. In fact, its superior performance
led to a new Department of Defense standard in 1996 and subsequently
to MIL-STD-3005 and NATO STANAG 4591, which operates at 2400,
1200, and 600 bps.
7.5 Frequency-Domain Coders
Although we have discussed a wide variety of digital speech coding
systems, we have really only focused on a few of the most important
examples that have current relevance. There is much room for varia-
tion within the general frameworks that we have identified. There is
one more class of coders that should be discussed. These are called
frequency-domain coders because they are based on the principle
of decomposing the speech signal into individual frequency bands.
Figure 7.13 depicts the general nature of this large class of coding
systems. On the far left of the diagram is a set (bank) of analysis band-
pass filters. Each of these is followed by a down-sampler by a factor
appropriate for the bandwidth of its bandpass input. On the far right
in the diagram is a set of “bandpass interpolators” where each interpo-
lator is composed of an up-sampler (i.e., raising the sampling rate by
the same factor as the corresponding down-sampling box) followed by
a bandpass filter similar to the analysis filter. If the filters are carefully
designed, and the outputs of the down-samplers are connected to the
corresponding inputs of the bandpass interpolators, then it is possi-
ble for the output ˆ x[n] to be virtually identical to the input x[n] [24].
Such perfect reconstruction filter bank systems are the basis for a wide-
ranging class of speech and audio coders called subband coders [25].
When the outputs of the filter bank are quantized by some sort of
quantizer, the output ˆ x[n] is not equal to the input, but by careful
7.5 Frequency-Domain Coders 133
Fig. 7.13 Subband coder and decoder for speech audio.
design of the quantizer the output can be indistinguishable from the
input perceptually. The goal with such coders is, of course, for the total
composite bit rate
23
to be as low as possible while maintaining high
quality.
In contrast to the coders that we have already discussed, which
incorporated the speech production model into the quantization pro-
cess, the filter bank structure incorporates important features of the
speech perception model. As discussed in Chapter 3, the basilar mem-
brane effectively performs a frequency analysis of the sound impinging
on the eardrum. The coupling between points on the basilar membrane
results in the masking effects that we mentioned in Chapter 3. This
masking was incorporated in some of the coding systems discussed so
far, but only in a rudimentary manner.
To see how subband coders work, suppose first that the individ-
ual channels are quantized independently. The quantizers can be any
quantization operator that preserves the waveform of the signal. Typ-
ically, adaptive PCM is used, but ADPCM could be used in princi-
ple. Because the down-sampled channel signals are full band at the
lower sampling rate, the quantization error affects all frequencies in
23
The total bit rate will be the sum of the products of the sampling rates of the down-
sampled channel signals times the number of bits allocated to each channel.
134 Digital Speech Coding
that band, but the important point is that no other bands are affected
by the errors in a given band. Just as important, a particular band
can be quantized according to its perceptual importance. Further-
more, bands with low energy can be encoded with correspondingly low
absolute error.
The simplest approach is to pre-assign a fixed number of bits to
each channel. In an early non-uniformly spaced five channel system,
Crochiere et al. [25] found that subband coders operating at 16 kbps
and 9.6 kbps were preferred by listeners to ADPCM coders operating
at 22 and 19 kbps, respectively.
Much more sophisticated quantization schemes are possible in the
configuration shown in Figure 7.13 if, as suggested by the dotted boxes,
the quantization of the channels is done jointly. For example, if the
channel bandwidths are all the same so that the output samples of the
down-samplers can be treated as a vector, vector quantization can be
used effectively [23].
Another approach is to allocate bits among the channels dynami-
cally according to a perceptual criterion. This is the basis for modern
audio coding standards such as the various MPEG standards. Such sys-
tems are based on the principles depicted in Figure 7.13. What is not
shown in that figure is additional analysis processing that is done to
determine how to allocate the bits among the channels. This typically
involves the computation of a fine-grained spectrum analysis using the
DFT. From this spectrum it is possible to determine which frequen-
cies will mask other frequencies, and on the basis of this a threshold of
audibility is determined. Using this audibility threshold, bits are allo-
cated among the channels in an iterative process so that as much of the
quantization error is inaudible as possible for the total bit budget. The
details of perceptual audio coding are presented in the recent textbook
by Spanias [120].
7.6 Evaluation of Coders
In this chapter, we have discussed a wide range of options for digital
coding of speech signals. These systems rely heavily on the techniques of
linear prediction, filter banks, and cepstrum analysis and also on models
7.6 Evaluation of Coders 135
for speech production and speech perception. All this knowledge and
technique is brought to bear on the problem of reducing the bit rate of
the speech representation while maintaining high quality reproduction
of the speech signal.
Throughout this chapter we have mentioned bit rates, complexity
of computation, processing delay, and quality as important practical
dimensions of the speech coding problem. Most coding schemes have
many operations and parameters that can be chosen to tradeoff among
the important factors. In general, increasing the bit rate will improve
quality of reproduction; however, it is important to note that for many
systems, increasing the bit rate indefinitely does not necessarily con-
tinue to improve quality. For example, increasing the bit rate of an
open-loop two-state-excitation LPC vocoder above about 2400 bps does
not improve quality very much, but lowering the bit rate causes notice-
able degradation. On the other hand, improved pitch detection and
source modeling can improve quality in an LPC vocoder as witnessed
by the success of the MELP system, but this generally would come
with an increase in computation and processing delay.
In the final analysis, the choice of a speech coder will depend on
constraints imposed by the application such as cost of coder/decoder,
available transmission capacity, robustness to transmission errors,
and quality requirements. Fortunately, there are many possibilities to
choose from.
8
Text-to-Speech Synthesis Methods
In this chapter, we discuss systems whose goal is to convert ordinary
text messages into intelligible and natural sounding synthetic speech
so as to transmit information from a machine to a human user. In
other words, we will be concerned with computer simulation of the
upper part of the speech chain in Figure 1.2. Such systems are often
referred to as text-to-speech synthesis (or TTS) systems, and their gen-
eral structure is illustrated in Figure 8.1. The input to the TTS sys-
tem is text and the output is synthetic speech. The two fundamental
processes performed by all TTS systems are text analysis (to deter-
mine the abstract underlying linguistic description of the speech) and
speech synthesis (to produce the speech sounds corresponding to the
text input).
Fig. 8.1 Block diagram of general TTS system.
136
8.1 Text Analysis 137
8.1 Text Analysis
The text analysis module of Figure 8.1 must determine three things
from the input text string, namely:
(1) pronunciation of the text string: the text analysis pro-
cess must decide on the set of phonemes that is to be spoken,
the degree of stress at various points in speaking, the into-
nation of the speech, and the duration of each of the sounds
in the utterance
(2) syntactic structure of the sentence to be spoken: the
text analysis process must determine where to place pauses,
what rate of speaking is most appropriate for the material
being spoken, and how much emphasis should be given to
individual words and phrases within the final spoken output
speech
(3) semantic focus and ambiguity resolution: the text anal-
ysis process must resolve homographs (words that are spelled
alike but can be pronounced differently, depending on con-
text), and also must use rules to determine word etymology
to decide on how best to pronounce names and foreign words
and phrases.
Figure 8.2 shows more detail on how text analysis is performed.
The input to the analysis is plain English text. The first stage of pro-
cessing does some basic text processing operations, including detecting
the structure of the document containing the text (e.g., email message
versus paragraph of text from an encyclopedia article), normalizing the
text (so as to determine how to pronounce words like proper names or
homographs with multiple pronunciations), and finally performing a
linguistic analysis to determine grammatical information about words
and phrases within the text. The basic text processing benefits from
an online dictionary of word pronunciations along with rules for deter-
mining word etymology. The output of the basic text processing step
is tagged text, where the tags denote the linguistic properties of the
words of the input text string.
138 Text-to-Speech Synthesis Methods
Fig. 8.2 Components of the text analysis process.
8.1.1 Document Structure Detection
The document structure detection module seeks to determine the loca-
tion of all punctuation marks in the text, and to decide their significance
with regard to the sentence and paragraph structure of the input text.
For example, an end of sentence marker is usually a period, ., a ques-
tion mark, ?, or an exclamation point, !. However this is not always the
case as in the sentence, “This car is 72.5 in. long” where there are two
periods, neither of which denote the end of the sentence.
8.1.2 Text Normalization
Text normalization methods handle a range of text problems that occur
in real applications of TTS systems, including how to handle abbrevi-
ations and acronyms as in the following sentences:
Example 1: “I live on Bourbon St. in St. Louis”
Example 2: “She worked for DEC in Maynard MA”
8.1 Text Analysis 139
where, in Example 1, the text “St.” is pronounced as street or saint,
depending on the context, and in Example 2 the acronym DEC can
be pronounced as either the word “deck” (the spoken acronym) or
the name of the company, i.e., Digital Equipment Corporation, but is
virtually never pronounced as the letter sequence “D E C.”
Other examples of text normalization include number strings like
“1920” which can be pronounced as the year “nineteen twenty” or the
number “one thousand, nine hundred, and twenty,” and dates, times
currency, account numbers, etc. Thus the string “$10.50” should be
pronounced as “ten dollars and fifty cents” rather than as a sequence
of characters.
One other important text normalization problem concerns the pro-
nunciation of proper names, especially those from languages other than
English.
8.1.3 Linguistic Analysis
The third step in the basic text processing block of Figure 8.2 is a
linguistic analysis of the input text, with the goal of determining, for
each word in the printed string, the following linguistic properties:
• the part of speech (POS) of the word
• the sense in which each word is used in the current context
• the location of phrases (or phrase groups) with a sentence (or
paragraph), i.e., where a pause in speaking might be appro-
priate
• the presence of anaphora (e.g., the use of a pronoun to refer
back to another word unit)
• the word (or words) on which emphasis are to be placed, for
prominence in the sentence
• the style of speaking, e.g., irate, emotional, relaxed, etc.
A conventional parser could be used as the basis of the linguistic anal-
ysis of the printed text, but typically a simple, shallow analysis is per-
formed since most linguistic parsers are very slow.
140 Text-to-Speech Synthesis Methods
8.1.4 Phonetic Analysis
Ultimately, the tagged text obtained from the basic text processing
block of a TTS system has to be converted to a sequence of tagged
phones which describe both the sounds to be produced as well as the
manner of speaking, both locally (emphasis) and globally (speaking
style). The phonetic analysis block of Figure 8.2 provides the process-
ing that enables the TTS system to perform this conversion, with the
help of a pronunciation dictionary. The way in which these steps are
performed is as follows.
8.1.5 Homograph Disambiguation
The homograph disambiguation operation must resolve the correct pro-
nunciation of each word in the input string that has more than one pro-
nunciation. The basis for this is the context in which the word occurs.
One simple example of homograph disambiguation is seen in the phrases
“an absent boy” versus the sentence “do you choose to absent yourself.”
In the first phrase the word “absent” is an adjective and the accent is
on the first syllable; in the second phrase the word “absent” is a verb
and the accent is on the second syllable.
8.1.6 Letter-to-Sound (LTS) Conversion
The second step of phonetic analysis is the process of grapheme-to-
phoneme conversion, namely conversion from the text to (marked)
speech sounds. Although there are a variety of ways of performing this
analysis, perhaps the most straightforward method is to rely on a stan-
dard pronunciation dictionary, along with a set of letter-to-sound rules
for words outside the dictionary.
Figure 8.3 shows the processing for a simple dictionary search for
word pronunciation. Each individual word in the text string is searched
independently. First a “whole word” search is initiated to see if the
printed word exists, in its entirety, in the word dictionary. If so, the
conversion to sounds is straightforward and the dictionary search begins
on the next word. If not, as is the case most often, the dictionary search
attempts to find affixes (both prefixes and suffixes) and strips them
8.1 Text Analysis 141
Fig. 8.3 Block diagram of dictionary search for proper word pronunciation.
from the word attempting to find the “root form” of the word, and
then does another “whole word” search. If the root form is not present
in the dictionary, a set of letter-to-sound rules is used to determine the
best pronunciation (usually based on etymology of the word) of the
root form of the word, again followed by the reattachment of stripped
out affixes (including the case of no stripped out affixes).
8.1.7 Prosodic Analysis
The last step in the text analysis system of Figure 8.2 is prosodic anal-
ysis which provides the speech synthesizer with the complete set of
synthesis controls, namely the sequence of speech sounds, their dura-
tions, and an associated pitch contour (variation of fundamental fre-
quency with time). The determination of the sequence of speech sounds
is mainly performed by the phonetic analysis step as outlined above.
The assignment of duration and pitch contours is done by a set of pitch
142 Text-to-Speech Synthesis Methods
and duration rules, along with a set of rules for assigning stress and
determining where appropriate pauses should be inserted so that the
local and global speaking rates appear to be natural.
8.2 Evolution of Speech Synthesis Systems
A summary of the progress in speech synthesis, over the period 1962–
1997, is given in Figure 8.4. This figure shows that there have been
3 generations of speech synthesis systems. During the first genera-
tion (between 1962 and 1977) formant synthesis of phonemes using
a terminal analog synthesizer was the dominant technology using
rules which related the phonetic decomposition of the sentence to
formant frequency contours. The synthesis suffered from poor intel-
ligibility and poor naturalness. The second generation of speech syn-
thesis methods (from 1977 to 1992) was based primarily on an LPC
representation of sub-word units such as diphones (half phones). By
carefully modeling and representing diphone units via LPC param-
eters, it was shown that good intelligibility synthetic speech could
be reliably obtained from text input by concatenating the appropri-
ate diphone units. Although the intelligibility improved dramatically
Fig. 8.4 Time line of progress in speech synthesis and TTS systems.
8.2 Evolution of Speech Synthesis Systems 143
over first generation formant synthesis, the naturalness of the syn-
thetic speech remained low due to the inability of single diphone units
to represent all possible combinations of sound using that diphone
unit. A detailed survey of progress in text-to-speech conversion up
to 1987 is given in the review paper by Klatt [65]. The synthesis
examples that accompanied that paper are available for listening at
http://www.cs.indiana.edu/rhythmsp/ASA/Contents.html.
The third generation of speech synthesis technology was the period
from 1992 to the present, in which the method of “unit selection syn-
thesis” was introduced and perfected, primarily by Sagisaka at ATR
Labs in Kyoto [108]. The resulting synthetic speech from this third
generation technology had good intelligibility and naturalness that
approached that of human-generated speech. We begin our discussion
of speech synthesis approaches with a review of early systems, and then
discuss unit selection methods of speech synthesis later in this chapter.
8.2.1 Early Speech Synthesis Approaches
Once the abstract underlying linguistic description of the text input has
been determined via the steps of Figure 8.2, the remaining (major) task
of TTS systems is to synthesize a speech waveform whose intelligibility
is very high (to make the speech useful as a means of communication
between a machine and a human), and whose naturalness is as close to
real speech as possible. Both tasks, namely attaining high intelligibility
along with reasonable naturalness, are difficult to achieve and depend
critically on three issues in the processing of the speech synthesizer
“backend” of Figure 8.1, namely:
(1) choice of synthesis units: including whole words, phones,
diphones, dyads, or syllables
(2) choice of synthesis parameters: including LPC features,
formants, waveform templates, articulatory parameters, sinu-
soidal parameters, etc.
(3) method of computation: including rule-based systems or
systems which rely on the concatenation of stored speech
units.
144 Text-to-Speech Synthesis Methods
8.2.2 Word Concatenation Synthesis
Perhaps the simplest approach to creating a speech utterance corre-
sponding to a given text string is to literally splice together prerecorded
words corresponding to the desired utterance. For greatest simplicity,
the words can be stored as sampled waveforms and simply concatenated
in the correct sequence. This approach generally produces intelligible,
but unnatural sounding speech, since it does not take into account the
“co-articulation” effects of producing phonemes in continuous speech,
and it does not provide either for the adjustment of phoneme dura-
tions or the imposition of a desired pitch variation across the utter-
ance. Words spoken in continuous speech sentences are generally much
shorter in duration than when spoken in isolation (often up to 50%
shorter) as illustrated in Figure 8.5. This figure shows wideband spec-
trograms for the sentence “This shirt is red” spoken as a sequence of
isolated words (with short, distinct pauses between words) as shown
at the top of Figure 8.5, and as a continuous utterance, as shown at
the bottom of Figure 8.5. It can be seen that, even for this trivial
Fig. 8.5 Wideband spectrograms of a sentence spoken as a sequence of isolated words (top
panel) and as a continuous speech utterance (bottom panel).
8.2 Evolution of Speech Synthesis Systems 145
example, the duration of the continuous sentence is on the order of half
that of the isolated word version, and further the formant tracks of the
continuous sentence do not look like a set of uniformly compressed for-
mant tracks from the individual words. Furthermore, the boundaries
between words in the upper plot are sharply defined, while they are
merged in the lower plot.
The word concatenation approach can be made more sophisticated
by storing the vocabulary words in a parametric form (formants, LPC
parameters, etc.) such as employed in the speech coders discussed in
Chapter 7 [102]. The rational for this is that the parametric represen-
tations, being more closely related to the model for speech produc-
tion, can be manipulated so as to blend the words together, shorten
them, and impose a desired pitch variation. This requires that the con-
trol parameters for all the words in the task vocabulary (as obtained
from a training set of words) be stored as representations of the words.
A special set of word concatenation rules is then used to create the
control signals for the synthesizer. Although such a synthesis system
would appear to be an attractive alternative for general purpose syn-
thesis of speech, in reality this type of synthesis is not a practical
approach.
1
There are many reasons for this, but a major problem is that there
are far too many words to store in a word catalog for word concatena-
tion synthesis to be practical except in highly restricted situations. For
example, there are about 1.7 million distinct surnames in the United
States and each of them would have to be spoken and stored for a gen-
eral word concatenation synthesis method. A second, equally important
limitation is that word-length segments of speech are simply the wrong
sub-units. As we will see, shorter units such as phonemes or diphones
are more suitable for synthesis.
Efforts to overcome the limitations of word concatenation followed
two separate paths. One approach was based on controlling the motions
of a physical model of the speech articulators based on the sequence of
phonemes from the text analysis. This requires sophisticated control
1
It should be obvious that whole word concatenation synthesis (from stored waveforms) is
also impractical for general purpose synthesis of speech.
146 Text-to-Speech Synthesis Methods
rules that are mostly derived empirically. From the vocal tract shapes
and sources produced by the articulatory model, control parameters
(e.g., formants and pitch) can be derived by applying the acoustic theory
of speech production and then used to control a synthesizer such those
used for speech coding. An alternative approach eschews the articulatory
model and proceeds directly to the computation of the control signals
for a source/system model (e.g., formant parameters, LPC parameters,
pitch period, etc.). Again, the rules (algorithm) for computing the control
parameters are mostly derived by empirical means.
8.2.3 Articulatory Methods of Synthesis
Articulatory models of human speech are based upon the application of
acoustic theory to physical models such as depicted in highly stylized
form in Figure 2.1 [22]. Such models were thought to be inherently
more natural than vocal tract analog models since:
• we could impose known and fairly well understood phys-
ical constraints on articular movements to create realistic
motions of the tongue, jaw, teeth, velum, etc.
• we could use X-ray data (MRI data today) to study the
motion of the articulators in the production of individual
speech sounds, thereby increasing our understanding of the
dynamics of speech production
• we could model smooth articulatory parameter motions
between sounds, either via direct methods (namely solving
the wave equation), or indirectly by converting articulatory
shapes to formants or LPC parameters
• we could highly constrain the motions of the articu-
latory parameters so that only natural motions would
occur, thereby potentially making the speech more natural
sounding.
What we have learned about articulatory modeling of speech is
that it requires a highly accurate model of the vocal cords and of the
vocal tract for the resulting synthetic speech quality to be considered
acceptable. It further requires rules for handling the dynamics of the
8.2 Evolution of Speech Synthesis Systems 147
articulator motion in the context of the sounds being produced. So
far we have been unable to learn all such rules, and thus, articula-
tory speech synthesis methods have not been found to be practical for
synthesizing speech of acceptable quality.
8.2.4 Terminal Analog Synthesis of Speech
The alternative to articulatory synthesis is called terminal analog speech
synthesis. In this approach, each sound of the language (phoneme) is
characterized by a source excitation function and an ideal vocal tract
model. Speech is produced by varying (in time) the excitation and the
vocal tract model control parameters at a rate commensurate with the
sounds being produced. This synthesis process has been called terminal
analog synthesis because it is based on a model (analog) of the human
vocal tract production of speech that seeks to produce a signal at its
output terminals that is equivalent to the signal produced by a human
talker.
2
The basis for terminal analog synthesis of speech is the
source/system model of speech production that we have used many
times in this text; namely an excitation source, e[n], and a transfer
function of the human vocal tract in the form of a rational system
function, i.e.,
H(z) =
S(z)
E(z)
=
B(z)
A(z)
=
b
0
+
q
¸
k=1
b
k
z
−k
1 −
p
¸
k=1
a
k
z
−k
, (8.1)
where S(z) is the z-transform of the output speech signal, s[n],
E(z) is the z-transform of the vocal tract excitation signal, e[n],
and ¦b
k
¦ = ¦b
0
, b
1
, b
2
, . . . , b
q
¦ and ¦a
k
¦ = ¦a
1
, a
2
, . . . , a
p
¦ are the (time-
varying) coefficients of the vocal tract filter. (See Figures 2.2 and 4.1.)
2
In the designation “terminal analog synthesis”, the terms analog and terminal result from
the historical context of early speech synthesis studies. This can be confusing since today,
“analog” implies “not digital” as well as an analogous thing. “Terminal” originally implied
the “output terminals” of an electronic analog (not digital) circuit or system that was an
analog of the human speech production system.
148 Text-to-Speech Synthesis Methods
For most practical speech synthesis systems, both the excitation signal
properties (P
0
and V/UV ) and the filter coefficients of (8.1) change
periodically so as to synthesize different phonemes.
The vocal tract representation of (8.1) can be implemented as a
speech synthesis system using a direct form implementation. However,
it has been shown that it is preferable to factor the numerator and
denominator polynomials into either a series of cascade (serial) reso-
nances, or into a parallel combination of resonances. It has also been
shown that an all-pole model (B(z) =constant) is most appropriate for
voiced (non-nasal) speech. For unvoiced speech a simpler model (with
one complex pole and one complex zero), implemented via a paral-
lel branch is adequate. Finally a fixed spectral compensation network
can be used to model the combined effects of glottal pulse shape and
radiation characteristics from the lips and mouth, based on two real
poles in the z-plane. A complete serial terminal analog speech syn-
thesis model, based on the above discussion, is shown in Figure 8.6
[95, 111].
The voiced speech (upper) branch includes an impulse generator
(controlled by a time-varying pitch period, P
0
), a time-varying voiced
signal gain, A
V
, and an all-pole discrete-time system that consists
of a cascade of 3 time-varying resonances (the first three formants,
F
1
, F
2
, F
3
), and one fixed resonance F
4
.
Fig. 8.6 Speech synthesizer based on a cascade/serial (formant) synthesis model.
8.3 Unit Selection Methods 149
The unvoiced speech (lower) branch includes a white noise
generator, a time-varying unvoiced signal gain, A
N
, and a reso-
nance/antiresonance system consisting of a time-varying pole (F
P
) and
a time-varying zero (F
Z
).
The voiced and unvoiced components are added and processed by
the fixed spectral compensation network to provide the final synthetic
speech output.
The resulting quality of the synthetic speech produced using a ter-
minal analog synthesizer of the type shown in Figure 8.6 is highly vari-
able with explicit model shortcomings due to the following:
• voiced fricatives are not handled properly since their mixed
excitation is not part of the model of Figure 8.6
• nasal sounds are not handled properly since nasal zeros are
not included in the model
• stop consonants are not handled properly since there is no
precise timing and control of the complex excitation signal
• use of a fixed pitch pulse shape, independent of the pitch
period, is inadequate and produces buzzy sounding voiced
speech
• the spectral compensation model is inaccurate and does not
work well for unvoiced sounds.
Many of the short comings of the model of Figure 8.6 are alleviated by
a more complex model proposed by Klatt [64]. However, even with a
more sophisticated synthesis model, it remains a very challenging task
to compute the synthesis parameters. Nevertheless Klatt’s Klattalk
system achieved adequate quality by 1983 to justify commercializa-
tion by the Digital Equipment Corporation as the DECtalk system.
Some DECtalk systems are still in operation today as legacy systems,
although the unit selection methods to be discussed in Section 8.3 now
provide superior quality in current applications.
8.3 Unit Selection Methods
The key idea of a concatenative TTS system, using unit selection
methods, is to use synthesis segments that are sections of prerecorded
150 Text-to-Speech Synthesis Methods
natural speech [31, 50, 108]. The word concatenation method discussed
in Section 8.2.2 is perhaps the simplest embodiment of this idea; how-
ever, as we discussed, shorter segments are required to achieve better
quality synthesis. The basic idea is that the more segments recorded,
annotated, and saved in the database, the better the potential quality
of the resulting speech synthesis. Ultimately, if an infinite number of
segments were recorded and saved, the resulting synthetic speech would
sound natural for virtually all possible synthesis tasks. Concatenative
speech synthesis systems, based on unit selection methods, are what is
conventionally known as “data driven” approaches since their perfor-
mance tends to get better the more data that is used for training the
system and selecting appropriate units.
In order to design and build a unit selection system based on using
recorded speech segments, several issues have to be resolved, including
the following:
(1) What speech units should be used as the basic synthesis
building blocks?
(2) How the synthesis units are selected (extracted) from natural
speech utterances?
(3) How the units are labeled for retrieval from a large database
of units?
(4) What signal representation should be used to represent the
units for storage and reproduction purposes?
(5) What signal processing methods can be used to spectrally
smooth the units (at unit junctures) and for prosody modi-
fication (pitch, duration, amplitude)?
We now attempt to answer each of these questions.
8.3.1 Choice of Concatenation Units
The units for unit selection can (in theory) be as large as words and
as small as phoneme units. Words are prohibitive since there are essen-
tially an infinite number of words in English. Subword units include
syllables (about 10,000 in English), phonemes (about 45 in English,
but they are highly context dependent), demi-syllables (about 2500
8.3 Unit Selection Methods 151
in English), and diphones (about 1500–2500 in English). The ideal
synthesis unit is context independent and easily concatenates with
other (appropriate) subword units [15]. Based on this criterion, the
most reasonable choice for unit selection synthesis is the set of diphone
units.
3
Before any synthesis can be done, it is necessary to prepare an inven-
tory of units (diphones). This requires significant effort if done manu-
ally. In any case, a large corpus of speech must be obtained from which
to extract the diphone units as waveform snippets. These units are then
represented in some compressed form for efficient storage. High quality
coding such as MPLP or CELP is used to limit compression artifacts.
At the final synthesis stage, the diphones are decoded into waveforms
for final merging, duration adjustment and pitch modification, all of
which take place on the time waveform.
8.3.2 From Text to Diphones
The synthesis procedure from subword units is straightforward, but far
from trivial [14]. Following the process of text analysis into phonemes
and prosody, the phoneme sequence is first converted to the appropriate
sequence of units from the inventory. For example, the phrase “I want”
would be converted to diphone units as follows:
Text Input: I want.
Phonemes: /#/ AY/ /W/ /AA/ /N/ /T/ /#/
Diphones: /# AY/ /AY-W/ /W-AA/ /AA-N/ /N-T/ /T- #/,
where the symbol # is used to represent silence (at the beginning and
end of each sentence or phrase).
The second step in online synthesis is to select the most appropri-
ate sequence of diphone units from the stored inventory. Since each
diphone unit occurs many times in the stored inventory, the selection
3
A diphone is simply the concatenation of two phonemes that are allowed by the language
constraints to be sequentially contiguous in natural speech. For example, a diphone based
upon the phonemes AY and W would be denoted AY−W where the − denotes the joining
of the two phonemes.
152 Text-to-Speech Synthesis Methods
of the best sequence of diphone units involves solving a dynamic pro-
gramming search for the sequence of units that minimizes a specified
cost function. The cost function generally is based on diphone matches
at each of the boundaries between diphones, where the diphone match
is defined in terms of spectral matching characteristics, pitch matching
characteristics, and possibly phase matching characteristics.
8.3.3 Unit Selection Synthesis
The “Unit Selection” problem is basically one of having a given set of
target features corresponding to the spoken text, and then automat-
ically finding the sequence of units in the database that most closely
match these features. This problem is illustrated in Figure 8.7 which
shows a target feature set corresponding to the sequence of sounds
(phonemes for this trivial example) /HH/ /EH/ /L/ /OW/ from the
word /hello/.
4
As shown, each of the phonemes in this word have mul-
tiple representations (units) in the inventory of sounds, having been
extracted from different phonemic environments. Hence there are many
versions of /HH/ and many versions of /EH/ etc., as illustrated in
Figure 8.7. The task of the unit selection module is to choose one of
each of the multiple representations of the sounds, with the goal being
to minimize the total perceptual distance between segments of the cho-
sen sequence, based on spectral, pitch and phase differences throughout
the sounds and especially at the boundary between sounds. By spec-
ifying costs associated with each unit, both globally across the unit,
HH
EH
L
OW
•••
H
H
EH
L
O
W
“Trained” Perceptual
Distance
HH
H
H
H
H
EH
E
H
E
H
E
H
E
H L
L
L
L
L
O
W
OW
O
W
Fig. 8.7 Illustration of basic process of unit selection. (After Dutoit [31].)
4
If diphones were used, the sequence would be /HH-EH/ /EH-L/ /L-OW/.
8.3 Unit Selection Methods 153
Fig. 8.8 Online unit selection based on a Viterbi search through a lattice of alternatives.
and locally at the unit boundaries with adjacent units, we can find
the sequence of units that best “join each other” in the sense of min-
imizing the accumulated distance across the sequence of units. This
optimal sequence (or equivalently the optimal path through the combi-
nation of all possible versions of each unit in the string) can be found
using a Viterbi search (dynamic programming) [38, 99]. This dynamic
programming search process is illustrated in Figure 8.8 for a 3 unit
search. The Viterbi search effectively computes the cost of every pos-
sible path through the lattice and determines the path with the low-
est total cost, where the transitional costs (the arcs) reflect the cost
of concatenation of a pair of units based on acoustic distances, and
the nodes represent the target costs based on the linguistic identity of
the unit.
Thus there are two costs associated with the Viterbi search, namely
a nodal cost based on the unit segmental distortion (USD) which is
defined as the difference between the desired spectral pattern of the
target (suitably defined) and that of the candidate unit, throughout
the unit, and the unit concatenative distortion (UCD) which is defined
as the spectral (and/or pitch and/or phase) discontinuity across the
boundaries of the concatenated units. By way of example, consider
a target context of the word “cart” with target phonemes /K/ /AH/
/R/ /T/, and a source context of the phoneme /AH/ obtained from the
154 Text-to-Speech Synthesis Methods
Fig. 8.9 Illustration of unit selection costs associated with a string of target units and a
presumptive string of selected units.
source word “want” with source phonemes /W/ /AH/ /N/ /T/. The
USD distance would be the cost (specified analytically) between the
sound /AH/ in the context /W/ /AH/ /N/ versus the desired sound
in the context /K/ /AH/ /R/.
Figure 8.9 illustrates concatenative synthesis for a given string of
target units (at the bottom of the figure) and a string of selected units
from the unit inventory (shown at the top of the figure). We focus
our attention on Target Unit t
j
and Selected Units θ
j
and θ
j+1
. Asso-
ciated with the match between units t
j
and θ
j
is a USD Unit Cost
and associated with the sequence of selected units, θ
j
and θ
j+1
, is a
UCD concatenation cost. The total cost of an arbitrary string of N
selected units, Θ = ¦θ
1
, θ
2
, . . . , θ
N
¦, and the string of N target units,
T = ¦t
1
, t
2
, . . . , t
N
¦ is defined as:
d(Θ, T) =
N
¸
j=1
d
u

j
, t
j
) +
N−1
¸
j=1
d
t

j
, θ
j+1
), (8.2)
where d
u

j
, t
j
) is the USD cost associated with matching target unit t
j
with selected unit θ
j
and d
t

j
, θ
j+1
) is the UCD cost associated with
concatenating the units θ
j
and θ
j+1
. It should be noted that when
selected units θ
j
and θ
j+1
come from the same source sentence and
are adjacent units, the UCD cost goes to zero, as this is as natural a
concatenation as can exist in the database. Further, it is noted that
generally there is a small overlap region between concatenated units.
The optimal path (corresponding to the optimal sequence of units) can
be efficiently computed using a standard Viterbi search ([38, 99]) in
8.3 Unit Selection Methods 155
which the computational demand scales linearly with both the number
of target and the number of concatenation units.
The concatenation cost between two units is essentially the spectral
(and/or phase and/or pitch) discontinuity across the boundary between
the units, and is defined as:
d
t

j
, θ
j+1
) =
p+2
¸
k=1
w
k
C
k

j
, θ
j+1
), (8.3)
where p is the size of the spectral feature vector (typically p = 12 mfcc
coefficients (mel-frequency cepstral coefficients as explained in Sec-
tion 5.6.3), often represented as a VQ codebook vector), and the extra
two features are log power and pitch. The weights, w
k
are chosen dur-
ing the unit selection inventory creation phase and are optimized using
a trial-and-error procedure. The concatenation cost essentially mea-
sures a spectral plus log energy plus pitch difference between the two
concatenated units at the boundary frames. Clearly the definition of
concatenation cost could be extended to more than a single boundary
frame. Also, as stated earlier, the concatenation cost is defined to be
zero (d
t

j
, θ
j+1
) = 0) whenever units θ
j
and θ
j+1
are consecutive (in
the database) since, by definition, there is no discontinuity in either
spectrum or pitch in this case. Although there are a variety of choices
for measuring the spectral/log energy/pitch discontinuity at the bound-
ary, a common cost function is the normalized mean-squared error in
the feature parameters, namely:
C
k

j
, θ
j+1
) =
[f
θ
j
k
(m) − f
θ
j+1
k
(1)]
2
σ
2
k
, (8.4)
where f
θ
j
k
(l) is the kth feature parameter of the lth frame of segment θ
j
,
m is the (normalized) duration of each segment, and σ
2
k
is the variance
of the kth feature vector component.
The USD or target costs are conceptually more difficult to under-
stand, and, in practice, more difficult to instantiate. The USD cost
associated with units θ
j
and t
j
is of the form:
d
u

j
, t
j
) =
q
¸
i=1
w
t
i
φ
i

T
i
(f
θ
j
i
), T
i
(f
t
j
i
)
¸
, (8.5)
156 Text-to-Speech Synthesis Methods
where q is the number of features that specify the unit θ
j
or t
j
,
w
t
i
, i = 1, 2, . . . , q is a trained set of target weights, and T
i
() can be
either a continuous function (for a set of features, f
i
, such as segmental
pitch, power or duration), or a set of integers (in the case of cate-
gorical features, f
i
, such as unit identity, phonetic class, position in
the syllable from which the unit was extracted, etc.). In the latter
case, φ
i
can be looked up in a table of distances. Otherwise, the local
distance function can be expressed as a simple quadratic distance of
the form
φ
i

T
i
(f
θ
j
i
), T
i
(f
t
j
i
)
¸
=

T
i
(f
θ
j
i
) − T
i
(f
t
j
i
)

2
. (8.6)
The training of the weights, w
t
i
, is done off-line. For each phoneme
in each phonetic class in the training speech database (which might be
the entire recorded inventory), each exemplar of each unit is treated
as a target and all others are treated as candidate units. Using this
training set, a least-squares system of linear equations can be derived
from which the weight vector can be solved. The details of the weight
training methods are described by Schroeter [114].
The final step in the unit selection synthesis, having chosen the opti-
mal sequence of units to match the target sentence, is to smooth/modify
the selected units at each of the boundary frames to better match
the spectra, pitch and phase at each unit boundary. Various smooth-
ing methods based on the concepts of time domain harmonic scaling
(TDHS) [76] and pitch synchronous overlap add (PSOLA) [20, 31, 82]
have been proposed and optimized for such smoothing/modification at
the boundaries between adjacent diphone units.
8.4 TTS Applications
Speech technology serves as a way of intelligently and efficiently
enabling humans to interact with machines with the goal of bring-
ing down cost of service for an existing service capability, or providing
new products and services that would be prohibitively expensive with-
out the automation provided by a viable speech processing interactive
8.5 TTS Future Needs 157
system. Examples of existing services where TTS enables significant
cost reduction are as follows:
• a dialog component for customer care applications
• a means for delivering text messages over an audio connection
(e.g., cellphone)
• a means of replacing expensive recorded Interactive Voice
Response system prompts (a service which is particularly
valuable when the prompts change often during the course
of the day, e.g., stock price quotations).
Similarly some examples of new products and services which are
enabled by a viable TTS technology are the following:
• location-based services, e.g., alerting you to locations of
restaurants, gas stations, stores in the vicinity of your current
location
• providing information in cars (e.g., driving directions, traffic
reports)
• unified messaging (e.g., reading e-mail and fax messages)
• voice portal providing voice access to web-based services
• e-commerce agents
• customized News, Stock Reports, Sports scores, etc.
• giving voice to small and embedded devices for reporting
information and alerts.
8.5 TTS Future Needs
Modern TTS systems have the capability of producing highly intelligi-
ble, and surprisingly natural speech utterances, so long as the utterance
is not too long, or too syntactically complicated, or too technical. The
biggest problem with most TTS systems is that they have no idea as
to how things should be said, but instead rely on the text analysis for
emphasis, prosody, phrasing and all so-called suprasegmental features
of the spoken utterance. The more TTS systems learn how to produce
context-sensitive pronunciations of words (and phrases), the more nat-
ural sounding these systems will become. Hence, by way of example,
158 Text-to-Speech Synthesis Methods
the utterance “I gave the book to John” has at least three different
semantic interpretations, each with different emphasis on words in the
utterance, i.e.,
I gave the book to John, i.e., not to Mary or Bob.
I gave the book to John, i.e., not the photos or the
apple.
I gave the book to John, i.e., I did it, not someone else.
The second future need of TTS systems is improvements in the unit
selection process so as to better capture the target cost for mismatch
between predicted unit specification (i.e., phoneme name, duration,
pitch, spectral properties) and actual features of a candidate recorded
unit. Also needed in the unit selection process is a better spectral dis-
tance measure that incorporates human perception measures so as to
find the best sequence of units for a given utterance.
Finally, better signal processing would enable improved compression
of the units database, thereby making the footprint of TTS systems
small enough to be usable in handheld and mobile devices.
9
Automatic Speech Recognition (ASR)
In this chapter, we examine the process of speech recognition by
machine, which is in essence the inverse of the text-to-speech problem.
The driving factor behind research in machine recognition of speech
has been the potentially huge payoff of providing services where humans
interact solely with machines, thereby eliminating the cost of live agents
and significantly reducing the cost of providing services. Interestingly,
as a side benefit, this process often provides users with a natural and
convenient way of accessing information and services.
9.1 The Problem of Automatic Speech Recognition
The goal of an ASR system is to accurately and efficiently convert
a speech signal into a text message transcription of the spoken words,
independent of the device used to record the speech (i.e., the transducer
or microphone), the speaker’s accent, or the acoustic environment in
which the speaker is located (e.g., quiet office, noisy room, outdoors).
That is, the ultimate goal, which has not yet been achieved, is to per-
form as well as a human listener.
159
160 Automatic Speech Recognition (ASR)
Fig. 9.1 Conceptual model of speech production and speech recognition processes.
A simple conceptual model of the speech generation and speech
recognition processes is given in Figure 9.1, which is a simplified version
of the speech chain shown in Figure 1.2. It is assumed that the speaker
intends to express some thought as part of a process of conversing
with another human or with a machine. To express that thought, the
speaker must compose a linguistically meaningful sentence, W, in the
form of a sequence of words (possibly with pauses and other acoustic
events such as uh’s, um’s, er’s etc.). Once the words are chosen, the
speaker sends appropriate control signals to the articulatory speech
organs which form a speech utterance whose sounds are those required
to speak the desired sentence, resulting in the speech waveform s[n]. We
refer to the process of creating the speech waveform from the speaker’s
intention as the Speaker Model since it reflects the speaker’s accent and
choice of words to express a given thought or request. The processing
steps of the Speech Recognizer are shown at the right side of Figure 9.1
and consist of an acoustic processor which analyzes the speech signal
and converts it into a set of acoustic (spectral, temporal) features, X,
which efficiently characterize the speech sounds, followed by a linguistic
decoding process which makes a best (maximum likelihood) estimate of
the words of the spoken sentence, resulting in the recognized sentence
ˆ
W. This is in essence a digital simulation of the lower part of the speech
chain diagram in Figure 1.2.
Figure 9.2 shows a more detailed block diagram of the overall speech
recognition system. The input speech signal, s[n], is converted to the
sequence of feature vectors, X = ¦x
1
, x
2
, . . . , x
T
¦, by the feature analy-
sis block (also denoted spectral analysis). The feature vectors are com-
puted on a frame-by-frame basis using the techniques discussed in the
earlier chapters. In particular, the mel frequency cepstrum coefficients
9.2 Building a Speech Recognition System 161
Fig. 9.2 Block diagram of an overall speech recognition system.
are widely used to represent the short-time spectral characteristics. The
pattern classification block (also denoted as the decoding and search
block) decodes the sequence of feature vectors into a symbolic repre-
sentation that is the maximum likelihood string,
ˆ
W that could have
produced the input sequence of feature vectors. The pattern recogni-
tion system uses a set of acoustic models (represented as hidden Markov
models) and a word lexicon to provide the acoustic match score for each
proposed string. Also, an N-gram language model is used to compute
a language model score for each proposed word string. The final block
in the process is a confidence scoring process (also denoted as an utter-
ance verification block), which is used to provide a confidence score for
each individual word in the recognized string. Each of the operations
in Figure 9.2 involves many details and, in some cases, extensive digital
computation. The remainder of this chapter is an attempt to give the
flavor of what is involved in each part of Figure 9.2.
9.2 Building a Speech Recognition System
The steps in building and evaluating a speech recognition system are
the following:
(1) choose the feature set and the associated signal processing
for representing the properties of the speech signal over time
162 Automatic Speech Recognition (ASR)
(2) choose the recognition task, including the recognition word
vocabulary (the lexicon), the basic speech sounds to repre-
sent the vocabulary (the speech units), the task syntax or
language model, and the task semantics (if any)
(3) train the set of speech acoustic and language models
(4) evaluate performance of the resulting speech recognition
system.
Each of these steps may involve many choices and can involve signifi-
cant research and development effort. Some of the important issues are
summarized in this section.
9.2.1 Recognition Feature Set
There is no “standard” set of features for speech recognition. Instead,
various combinations of acoustic, articulatory, and auditory features
have been utilized in a range of speech recognition systems. The most
popular acoustic features have been the (LPC-derived) mel-frequency
cepstrum coefficients and their derivatives.
A block diagram of the signal processing used in most modern
large vocabulary speech recognition systems is shown in Figure 9.3.
The analog speech signal is sampled and quantized at rates between
8000 up to 20,000 samples/s. A first order (highpass) pre-emphasis net-
work (1 − αz
−1
) is used to compensate for the speech spectral falloff
at higher frequencies and approximates the inverse to the mouth trans-
mission frequency response. The pre-emphasized signal is next blocked
into frames of N samples, with adjacent frames spaced M samples
apart. Typical values for N and M correspond to frames of duration
15–40 ms, with frame shifts of 10 ms being most common; hence adja-
cent frames overlap by 5–30 ms depending on the chosen values of N
and M. A Hamming window is applied to each frame prior to spec-
tral analysis using either standard spectral analysis or LPC methods.
Following (optionally used) simple noise removal methods, the spec-
tral coefficients are normalized and converted to mel-frequency cep-
stral coefficients via standard analysis methods of the type discussed
in Chapter 5 [27]. Some type of cepstral bias removal is often used
prior to calculation of the first and second order cepstral derivatives.
9.2 Building a Speech Recognition System 163
Fig. 9.3 Block diagram of feature extraction process for feature vector consisting of mfcc
coefficients and their first and second derivatives.
Typically the resulting feature vector is the set of cepstral coefficients,
and their first- and second-order derivatives. It is typical to use about
13 mfcc coefficients, 13 first-order cepstral derivative coefficients, and
13 second-order derivative cepstral coefficients, making a D = 39 size
feature vector.
9.2.2 The Recognition Task
Recognition tasks vary from simple word and phrase recognition sys-
tems to large vocabulary conversational interfaces to machines. For
example, using a digits vocabulary, the task could be recognition of a
string of digits that forms a telephone number, or identification code,
or a highly constrained sequence of digits that form a password.
9.2.3 Recognition Training
There are two aspects to training models for speech recognition, namely
acoustic model training and language model training. Acoustic model
164 Automatic Speech Recognition (ASR)
training requires recording each of the model units (whole words,
phonemes) in as many contexts as possible so that the statistical
learning method can create accurate distributions for each of the
model states. Acoustic training relies on accurately labeled sequences of
speech utterances which are segmented according to the transcription,
so that training of the acoustic models first involves segmenting the
spoken strings into recognition model units (via either a Baum–Welch
[10, 11] or Viterbi alignment method), and then using the segmented
utterances to simultaneously build model distributions for each state of
the vocabulary unit models. The resulting statistical models form the
basis for the pattern recognition operations at the heart of the ASR
system. As discussed in Section 9.3, concepts such as Viterbi search are
employed in the pattern recognition process as well as in training.
Language model training requires a sequence of text strings that
reflect the syntax of spoken utterances for the task at hand. Generally
such text training sets are created automatically (based on a model of
grammar for the recognition task) or by using existing text sources,
such as magazine and newspaper articles, or closed caption transcripts
of television news broadcasts, etc. Other times, training sets for lan-
guage models can be created from databases, e.g., valid strings of tele-
phone numbers can be created from existing telephone directories.
9.2.4 Testing and Performance Evaluation
In order to improve the performance of any speech recognition system,
there must be a reliable and statistically significant way of evaluating
recognition system performance based on an independent test set of
labeled utterances. Typically we measure word error rate and sentence
(or task) error rate as a measure of recognizer performance. A brief
summary of performance evaluations across a range of ASR applica-
tions is given in Section 9.4.
9.3 The Decision Processes in ASR
The heart of any automatic speech recognition system is the pattern
classification and decision operations. In this section, we shall give a
brief introduction to these important topics.
9.3 The Decision Processes in ASR 165
9.3.1 Mathematical Formulation of the ASR Problem
The problem of automatic speech recognition is represented as a statis-
tical decision problem. Specifically it is formulated as a Bayes maximum
a posteriori probability (MAP) decision process where we seek to find
the word string
ˆ
W (in the task language) that maximizes the a pos-
teriori probability P(W[X) of that string, given the measured feature
vector, X, i.e.,
ˆ
W = argmax
W
P(W[X). (9.1)
Using Bayes’ rule we can rewrite (9.1) in the form:
ˆ
W = argmax
W
P(X[W)P(W)
P(X)
. (9.2)
Equation (9.2) shows that the calculation of the a posteriori probability
is decomposed into two terms, one that defines the a priori probability
of the word sequence, W, namely P(W), and the other that defines the
likelihood that the word string, W, produced the feature vector, X,
namely P(X[W). For all future calculations we disregard the denom-
inator term, P(X), since it is independent of the word sequence W
which is being optimized. The term P(X[W) is known as the “acoustic
model” and is generally denoted as P
A
(X[W) to emphasize the acous-
tic nature of this term. The term P(W) is known as the “language
model” and is generally denoted as P
L
(W) to emphasize the linguistic
nature of this term. The probabilities associated with P
A
(X[W) and
P
L
(W) are estimated or learned from a set of training data that have
been labeled by a knowledge source, usually a human expert, where the
training set is as large as reasonably possible. The recognition decoding
process of (9.2) is often written in the form of a 3-step process, i.e.,
ˆ
W = argmax
W
. .. .
Step 3
P
A
(X[W)
. .. .
Step 1
P
L
(W)
. .. .
Step 2
, (9.3)
where Step 1 is the computation of the probability associated with the
acoustic model of the speech sounds in the sentence W, Step 2 is the
computation of the probability associated with the linguistic model of
the words in the utterance, and Step 3 is the computation associated
166 Automatic Speech Recognition (ASR)
with the search through all valid sentences in the task language for the
maximum likelihood sentence.
In order to be more explicit about the signal processing and com-
putations associated with each of the three steps of (9.3) we need to
more explicit about the relationship between the feature vector, X, and
the word sequence W. As discussed above, the feature vector, X, is a
sequence of acoustic observations corresponding to each of T frames of
the speech, of the form:
X = ¦x
1
, x
2
, . . . , x
T
¦, (9.4)
where the speech signal duration is T frames (i.e., T times the frame
shift in ms) and each frame, x
t
, t = 1, 2, . . . , T is an acoustic feature
vector of the form:
x
t
= (x
t1
, x
t2
, . . . , x
tD
) (9.5)
that characterizes the spectral/temporal properties of the speech signal
at time t and D is the number of acoustic features in each frame.
Similarly we can express the optimally decoded word sequence, W, as:
W = w
1
, w
2
, . . . , w
M
, (9.6)
where there are assumed to be exactly M words in the decoded string.
9.3.2 The Hidden Markov Model
The most widely used method of building acoustic models (for both
phonemes and words) is the use of a statistical characterization known
as Hidden Markov Models (HMMs) [33, 69, 97, 98]. Figure 9.4 shows
a simple Q = 5-state HMM for modeling a whole word. Each HMM
state is characterized by a mixture density Gaussian distribution that
characterizes the statistical behavior of the feature vectors within the
states of the model [61, 62]. In addition to the statistical feature densi-
ties within states, the HMM is also characterized by an explicit set of
state transitions, a
ij
, which specify the probability of making a tran-
sition from state i to state j at each frame, thereby defining the time
sequence of the feature vectors over the duration of the word. Usually
the self-transitions, a
ii
are large (close to 1.0), and the jump transitions,
a
12
, a
23
, a
34
, a
45
, in the model, are small (close to 0).
9.3 The Decision Processes in ASR 167
Fig. 9.4 Word-based, left-to-right, HMM with 5 states.
The complete HMM characterization of a Q-state word model (or a
sub-word unit like a phoneme model) is generally written as λ(A, B, π)
with state transition matrix A = ¦a
ij
, 1 ≤ i, j ≤ Q¦, state observation
probability density, B = ¦b
j
(x
t
), 1 ≤ j ≤ Q¦, and initial state distribu-
tion, π = ¦π
i
, 1 ≤ i ≤ Q¦ with π
1
set to 1 for the “left-to-right” models
of the type shown in Figure 9.4.
In order to train the HMM (i.e., learn the optimal model param-
eters) for each word (or sub-word) unit, a labeled training set of sen-
tences (transcribed into words and sub-word units) is used to guide
an efficient training procedure known as the Baum–Welch algorithm
[10, 11].
1
This algorithm aligns each of the various words (or sub-
word units) with the spoken inputs and then estimates the appro-
priate means, covariances and mixture gains for the distributions in
each model state. The Baum–Welch method is a hill-climbing algo-
rithm and is iterated until a stable alignment of models and speech
is obtained. The details of the Baum–Welch procedure are beyond
the scope of this chapter but can be found in several references on
Speech Recognition methods [49, 99]. The heart of the training pro-
cedure for re-estimating HMM model parameters using the Baum–
Welch procedure is shown in Figure 9.5. An initial HMM model is
used to begin the training process. The initial model can be ran-
domly chosen or selected based on a priori knowledge of the model
1
The Baum–Welch algorithm is also widely referred to as the forward–backward method.
168 Automatic Speech Recognition (ASR)
Fig. 9.5 The Baum–Welch training procedure based on a given training set of utterances.
parameters. The iteration loop is a simple updating procedure for com-
puting the forward and backward model probabilities based on an input
speech database (the training set of utterances) and then optimizing
the model parameters to give an updated HMM. This process is iter-
ated until no further improvement in probabilities occurs with each new
iteration.
It is a simple matter to go from the HMM for a whole word,
as shown in Figure 9.4, to an HMM for a sub-word unit (such as a
phoneme) as shown in Figure 9.6. This simple 3-state HMM is a basic
sub-word unit model with an initial state representing the statistical
characteristics at the beginning of a sound, a middle state representing
the heart of the sound, and an ending state representing the spec-
tral characteristics at the end of the sound. A word model is made
by concatenating the appropriate sub-word HMMs, as illustrated in
Fig. 9.6 Sub-word-based HMM with 3 states.
9.3 The Decision Processes in ASR 169
Fig. 9.7 Word-based HMM for the word /is/ created by concatenating 3-state subword
models for the sub-word units /ih/ and /z/.
Figure 9.7, which concatenates the 3-state HMM for the sound /IH/
with the 3-state HMM for the sound /Z/, giving the word model for
the word “is” (pronounced as /IH Z/). In general the composition of
a word (from sub-word units) is specified in a word lexicon or dictio-
nary; however once the word model has been built it can be used much
the same as whole word models for training and for evaluating word
strings for maximizing the likelihood as part of the speech recognition
process.
We are now ready to define the procedure for aligning a sequence
of M word models, w
1
, w
w
, . . . , w
M
with a sequence of feature vectors,
X = ¦x
1
, x
2
, . . . , x
T
¦. The resulting alignment procedure is illustrated
in Figure 9.8. We see the sequence of feature vectors along the horizon-
tal axis and the concatenated sequence of word states along the vertical
axis. An optimal alignment procedure determines the exact best match-
ing sequence between word model states and feature vectors such that
the first feature vector, x
1
, aligns with the first state in the first word
model, and the last feature vector, x
T
, aligns with the last state in the
Mth word model. (For simplicity we show each word model as a 5-state
HMM in Figure 9.8, but clearly the alignment procedure works for any
size model for any word, subject to the constraint that the total num-
ber of feature vectors, T, exceeds the total number of model states, so
that every state has at least a single feature vector associated with that
state.) The procedure for obtaining the best alignment between feature
vectors and model states is based on either using the Baum–Welch sta-
tistical alignment procedure (in which we evaluate the probability of
every alignment path and add them up to determine the probability of
the word string), or a Viterbi alignment procedure [38, 132] for which
170 Automatic Speech Recognition (ASR)
Fig. 9.8 Alignment of concatenated HMM word models with acoustic feature vectors based
on either a Baum–Welch or Viterbi alignment procedure.
we determine the single best alignment path and use the probability
score along that path as the probability measure for the current word
string. The utility of the alignment procedure of Figure 9.8 is based on
the ease of evaluating the probability of any alignment path using the
Baum–Welch or Viterbi procedures.
We now return to the mathematical formulation of the ASR prob-
lem and examine in more detail the three steps in the decoding
Equation (9.3).
9.3.3 Step 1 — Acoustic Modeling
The function of the acoustic modeling step (Step 1) is to assign prob-
abilities to the acoustic realizations of a sequence of words, given the
observed acoustic vectors, i.e., we need to compute the probability that
the acoustic vector sequence X = ¦x
1
, x
2
, . . . , x
T
¦ came from the word
sequence W = w
1
, w
2
, . . . , w
M
(assuming each word is represented as an
HMM) and perform this computation for all possible word sequences.
9.3 The Decision Processes in ASR 171
This calculation can be expressed as:
P
A
(X[W) = P
A
(¦x
1
, x
2
, . . . , x
T
¦[w
1
, w
2
, . . . , w
M
). (9.7)
If we make the assumption that each frame, x
t
, is aligned with
word i(t) and HMM model state j(t) via the function w
i(t)
j(t)
and if we
assume that each frame is independent of every other frame, we can
express (9.7) as the product
P
A
(X[W) =
T
¸
t=1
P
A

x
t
[w
i(t)
j(t)

, (9.8)
where we associate each frame of X with a unique word and state,
w
i(t)
j(t)
, in the word sequence. Further, we calculate the local probability
P
A

x
t
[w
i(t)
j(t)

given that we know the word from which frame t came.
The process of assigning individual speech frames to the appropriate
word model in an utterance is based on an optimal alignment process
between the concatenated sequence of word models and the sequence
of feature vectors of the spoken input utterance being recognized. This
alignment process is illustrated in Figure 9.9 which shows the set of
T feature vectors (frames) along the horizontal axis, and the set of M
words (and word model states) along the vertical axis. The optimal
segmentation of these feature vectors (frames) into the M words is
shown by the sequence of boxes, each of which corresponds to one of
the words in the utterance and its set of optimally matching feature
vectors.
We assume that each word model is further decomposed into a set
of states which reflect the changing statistical properties of the fea-
ture vectors over time for the duration of the word. We assume that
each word is represented by an N-state HMM model, and we denote
the states as S
j
, j = 1, 2, . . . , N. Within each state of each word there
is a probability density that characterizes the statistical properties of
the feature vectors in that state. We have seen in the previous section
that the probability density of each state, and for each word, is learned
during a training phase of the recognizer. Using a mixture of Gaussian
densities to characterize the statistical distribution of the feature vec-
tors in each state, j, of the word model, which we denote as b
j
(x
t
),
172 Automatic Speech Recognition (ASR)
Fig. 9.9 Illustration of time alignment process between unknown utterance feature vectors
and set of M concatenated word models.
we get a state-based probability density of the form:
b
j
(x
t
) =
K
¸
k=1
c
jk
N[x
t
, µ
jk
, U
jk
], (9.9)
where K is the number of mixture components in the density function,
c
jk
is the weight of the kth mixture component in state j, with the
constraint c
jk
≥ 0, and N is a Gaussian density function with mean
vector, µ
jk
, for mixture k for state j, and covariance matrix, U
jk
, for
mixture k in state j. The density constraints are:
K
¸
k=1
c
jk
= 1, 1 ≤ j ≤ N (9.10)


−∞
b
j
(x
t
)dx
t
= 1, 1 ≤ j ≤ N. (9.11)
We now return to the issue of the calculation of the probability of
frame x
t
being associated with the j(t)th state of the i(t)th word in
9.3 The Decision Processes in ASR 173
the utterance, P
A
(x
t
[w
i(t)
j(t)
), which is calculated as
P
A

x
t
[w
i(t)
j(t)

= b
i(t)
j(t)
(x
t
), (9.12)
The computation of (9.12) is incomplete since we have ignored the
computation of the probability associated with the links between word
states, and we have also not specified how to determine the within-word
state, j, in the alignment between a given word and a set of feature
vectors corresponding to that word. We come back to these issues later
in this section.
The key point is that we assign probabilities to acoustic realizations
of a sequence of words by using hidden Markov models of the acoustic
feature vectors within words. Using an independent (and orthograph-
ically labeled) set of training data, we “train the system” and learn
the parameters of the best acoustic models for each word (or more
specifically for each sound that comprises each word). The parameters,
according to the mixture model of (9.9) are, for each state of the model,
the mixture weights, the mean vectors, and the covariance matrix.
Although we have been discussing acoustic models for whole words,
it should be clear that for any reasonable size speech recognition task,
it is impractical to create a separate acoustic model for every possible
word in the vocabulary since each word would have to be spoken in
every possible context in order to build a statistically reliable model
of the density functions of (9.9). Even for modest size vocabularies
of about 1000 words, the amount of training data required for word
models is excessive.
The alternative to word models is to build acoustic-phonetic models
for the 40 or so phonemes in the English language and construct the
model for a word by concatenating (stringing together sequentially)
the models for the constituent phones in the word (as represented in a
word dictionary or lexicon). The use of such sub-word acoustic-phonetic
models poses no real difficulties in either training or when used to
build up word models and hence is the most widely used representation
for building word models in a speech recognition system. State-of-the-
art systems use context-dependent phone models as the basic units of
recognition [99].
174 Automatic Speech Recognition (ASR)
9.3.4 Step 2 — The Language Model
The language model assigns probabilities to sequences of words, based
on the likelihood of that sequence of words occurring in the context
of the task being performed by the speech recognition system. Hence
the probability of the text string W =“Call home” for a telephone
number identification task is zero since that string makes no sense for
the specified task. There are many ways of building Language Models
for specific tasks, including:
(1) statistical training from text databases transcribed from
task-specific dialogs (a learning procedure)
(2) rule-based learning of the formal grammar associated with
the task
(3) enumerating, by hand, all valid text strings in the language
and assigning appropriate probability scores to each string.
The purpose of the language model, or grammar, is to enable the
computation of the a priori probability, P
L
(W), of a word string, W,
consistent with the recognition task [59, 60, 106]. Perhaps the most
popular way of constructing the language model is through the use
of a statistical N-gram word grammar that is estimated from a large
training set of text utterances, either from the task at hand or from a
generic database with applicability to a wide range of tasks. We now
describe the way in which such a language model is built.
Assume we have a large text training set of word-labeled utterances.
(Such databases could include millions or even tens of millions of text
sentences.) For every sentence in the training set, we have a text file
that identifies the words in that sentence. If we make the assumption
that the probability of a word in a sentence is conditioned on only
the previous N − 1 words, we have the basis for an N-gram language
model. Thus we assume we can write the probability of the sentence
W, according to an N-gram language model, as
P
L
(W) = P
L
(w
1
, w
2
, . . . , w
M
) (9.13)
=
M
¸
n=1
P
L
(w
n
[w
n−1
, w
n−2
, . . . , w
n−N+1
), (9.14)
9.3 The Decision Processes in ASR 175
where the probability of a word occurring in the sentence only depends
on the previous N − 1 words and we estimate this probability by
counting the relative frequencies of N-tuples of words in the train-
ing set. Thus, for example, to estimate word “trigram” probabilities
(i.e., the probability that a word w
n
was preceded by the pair of words
(w
n−1
, w
n−2
)), we compute this quantity as
P(w
n
[w
n−1
, w
n−2
) =
C(w
n−2
, w
n−1
, w
n
)
C(w
n−2
, w
n−1
)
, (9.15)
where C(w
n−2
, w
n−1
, w
n
) is the frequency count of the word triplet
(i.e., the trigram of words) consisting of (w
n−2
, w
n−1
, w
n
) as it occurs
in the text training set, and C(w
n−2
, w
n−1
) is the frequency count of
the word doublet (i.e., bigram of words) (w
n−2
, w
n−1
) as it occurs in
the text training set.
9.3.5 Step 3 — The Search Problem
The third step in the Bayesian approach to automatic speech recogni-
tion is to search the space of all valid word sequences from the language
model, to find the one with the maximum likelihood of having been spo-
ken. The key problem is that the potential size of the search space can
be astronomically large (for large vocabularies and high average word
branching factor language models), thereby taking inordinate amounts
of computing power to solve by heuristic methods. Fortunately, through
the use of methods from the field of Finite State Automata Theory,
Finite State Network (FSN) methods have evolved that reduce the
computational burden by orders of magnitude, thereby enabling exact
maximum likelihood solutions in computationally feasible times, even
for very large speech recognition problems [81].
The basic concept of a finite state network transducer is illustrated
in Figure 9.10 which shows a word pronunciation network for the word
/data/. Each arc in the state diagram corresponds to a phoneme in the
word pronunciation network, and the weight is an estimate of the proba-
bility that the arc is utilized in the pronunciation of the word in context.
We see that for the word /data/ there are four total pronunciations,
176 Automatic Speech Recognition (ASR)
Fig. 9.10 Word pronunciation transducer for four pronunciations of the word /data/.
(After Mohri [81].)
namely (along with their (estimated) pronunciation probabilities):
(1) /D/ /EY/ /D/ /AX/ — probability of 0.32
(2) /D/ /EY/ /T/ /AX/ — probability of 0.08
(3) /D/ /AE/ /D/ /AX/ — probability of 0.48
(4) /D/ /AE/ /T/ /AX/ — probability of 0.12.
The combined FSN of the 4 pronunciations is a lot more efficient than
using 4 separate enumerations of the word since all the arcs are shared
among the 4 pronunciations and the total computation for the full FSN
for the word /data/ is close to 1/4 the computation of the 4 variants
of the same word.
We can continue the process of creating efficient FSNs for each
word in the task vocabulary (the speech dictionary or lexicon), and
then combine word FSNs into sentence FSNs using the appropriate
language model. Further, we can carry the process down to the level of
HMM phones and HMM states, making the process even more efficient.
Ultimately we can compile a very large network of model states, model
phones, model words, and even model phrases, into a much smaller
network via the method of weighted finite state transducers (WFST),
which combine the various representations of speech and language and
optimize the resulting network to minimize the number of search states
(and, equivalently, thereby minimize the amount of duplicate computa-
tion). A simple example of such a WFST network optimization is given
in Figure 9.11 [81].
Using the techniques of network combination (which include net-
work composition, determinization, minimization, and weight push-
ing) and network optimization, the WFST uses a unified mathematical
9.4 Representative Recognition Performance 177
Fig. 9.11 Use of WFSTs to compile a set of FSNs into a single optimized network to
minimize redundancy in the network. (After Mohri [81].)
framework to efficiently compile a large network into a minimal rep-
resentation that is readily searched using standard Viterbi decoding
methods [38]. Using these methods, an unoptimized network with 10
22
states (the result of the cross product of model states, model phones,
model words, and model phrases) was able to be compiled down to
a mathematically equivalent model with 10
8
states that was readily
searched for the optimum word string with no loss of performance or
word accuracy.
9.4 Representative Recognition Performance
The block diagram of Figure 9.1 represents a wide range of possibil-
ities for the implementation of automatic speech recognition. In Sec-
tion 9.2.1, we suggest how the techniques of digital speech analysis
discussed in Chapters 4–6 can be applied to extract a sequence of fea-
ture vectors from the speech signal, and in Section 9.3 we describe the
most widely used statistical pattern recognition techniques that are
employed for mapping the sequence of feature vectors into a sequence
of symbols or words.
Many variations on this general theme have been investigated over
the past 30 years or more, and as we mentioned before, testing and
evaluation is a major part of speech recognition research and devel-
opment. Table 9.1 gives a summary of a range of the performance of
some speech recognition and natural language understanding systems
178 Automatic Speech Recognition (ASR)
Table 9.1 Word error rates for a range of speech recognition systems.
Corpus Type of Vocabulary Word
speech size error rate (%)
Connected digit strings
(TI Database)
Spontaneous 11 (0–9, oh) 0.3
Connected digit strings
(AT&T Mall
Recordings)
Spontaneous 11 (0–9, oh) 2.0
Connected digit strings
(AT&T HMIHY
c
)
Conversational 11 (0–9, oh) 5.0
Resource management
(RM)
Read speech 1000 2.0
Airline travel information
system (ATIS)
Spontaneous 2500 2.5
North American business
(NAB & WSJ)
Read text 64,000 6.6
Broadcast News Narrated News 210,000 ≈15
Switchboard Telephone 45,000 ≈27
conversation
Call-home Telephone 28,000 ≈35
conversation
that have been developed so far. This table covers a range of vocabulary
sizes, speaking styles, and application contexts [92].
It can be seen that for a vocabulary of 11 digits, the word error
rates are very low (0.3% for a very clean recording environment for
the TI (Texas Instruments) connected digits database [68]), but when
the digit strings are spoken in a noisy shopping mall environment the
word error rate rises to 2.0% and when embedded within conversational
speech (the AT&T HMIHY (How May I Help You) c (system) the word
error rate increases significantly to 5.0%, showing the lack of robustness
of the recognition system to noise and other background disturbances
[44]. Table 9.1 also shows the word error rates for a range of DARPA
tasks ranging from
• read speech of commands and informational requests about
a naval ships database (the resource management system or
RM) with a 1000 word vocabulary and a word error rate of
2.0%
• spontaneous speech input for booking airlines travel (the air-
line travel information system, or ATIS [133]) with a 2500
word vocabulary and a word error rate of 2.5%
9.5 Challenges in ASR Technology 179
• read text from a range of business magazines and newspa-
pers (the North American business task, or NAB) with a
vocabulary of 64,000 words and a word error rate of 6.6%
• narrated news broadcasts from a range of TV news providers
like CNBC, (the broadcast news task) with a 210,000 word
vocabulary and a word error rate of about 15%
• recorded live telephone conversations between two unrelated
individuals (the switchboard task [42]) with a vocabulary of
45,000 words and a word error rate of about 27%, and a
separate task for live telephone conversations between two
family members (the call home task) with a vocabulary of
28,000 words and a word error rate of about 35%.
9.5 Challenges in ASR Technology
So far, ASR systems fall far short of human speech perception in all
but the simplest, most constrained tasks. Before ASR systems become
ubiquitous in society, many improvements will be required in both
system performance and operational performance. In the system area
we need large improvements in accuracy, efficiency, and robustness in
order to utilize the technology for a wide range of tasks, on a wide
range of processors, and under a wide range of operating conditions.
In the operational area we need better methods of detecting when a
person is speaking to a machine and isolating the spoken input from
the background, we need to be able to handle users talking over the
voice prompts (so-called barge-in conditions), we need more reliable
and accurate utterance rejection methods so we can be sure that a word
needs to be repeated when poorly recognized the first time, and finally
we need better methods of confidence scoring of words, phrases, and
even sentences so as to maintain an intelligent dialog with a customer.
Conclusion
We have attempted to provide the reader with a broad overview of
the field of digital speech processing and to give some idea as to the
remarkable progress that has been achieved over the past 4–5 decades.
Digital speech processing systems have permeated society in the form of
cellular speech coders, synthesized speech response systems, and speech
recognition and understanding systems that handle a wide range of
requests about airline flights, stock price quotations, specialized help
desks etc.
There remain many difficult problems yet to be solved before digital
speech processing will be considered a mature science and technology.
Our basic understanding of the human articulatory system and how the
various muscular controls come together in the production of speech is
rudimentary at best, and our understanding of the processing of speech
in the human brain is at an even lower level.
In spite of our shortfalls in understanding, we have been able to cre-
ate remarkable speech processing systems whose performance increases
at a steady pace. A firm understanding of the fundamentals of acous-
tics, linguistics, signal processing, and perception provide the tools for
180
Conclusion 181
building systems that work and can be used by the general public. As
we increase our basic understanding of speech, the application systems
will only improve and the pervasiveness of speech processing in our
daily lives will increase dramatically, with the end result of improving
the productivity in our work and home environments.
Acknowledgments
We wish to thank Professor Robert Gray, Editor-in-Chief of Now Pub-
lisher’s Foundations and Trends in Signal Processing, for inviting us
to prepare this text. His patience, technical advice, and editorial skill
were crucial at every stage of the writing. We also wish to thank
Abeer Alwan, Yariv Ephraim, Luciana Ferrer, Sadaoki Furui, and Tom
Quatieri for their detailed and perceptive comments, which greatly
improved the final result. Of course, we are responsible for any weak-
nesses or inaccuracies that remain in this text.
182
References
[1] J. B. Allen and L. R. Rabiner, “A unified theory of short-time spectrum
analysis and synthesis,” Proceedings of IEEE, vol. 65, no. 11, pp. 1558–1564,
November 1977.
[2] B. S. Atal, “Predictive coding of speech at low bit rates,” IEEE Transactions
on Communications, vol. COM-30, no. 4, pp. 600–614, April 1982.
[3] B. S. Atal and S. L. Hanauer, “Speech analysis and synthesis by linear predic-
tion of the speech wave,” Journal of the Acoustical Society of America, vol. 50,
pp. 561–580, 1971.
[4] B. S. Atal and J. Remde, “A new model of LPC exitation for producing
natural-sounding speech at low bit rates,” Proceedings of IEEE ICASSP,
pp. 614–617, 1982.
[5] B. S. Atal and M. R. Schroeder, “Adaptive predictive coding of speech sig-
nals,” Bell System Technical Journal, vol. 49, pp. 1973–1986, October 1970.
[6] B. S. Atal and M. R. Schroeder, “Predictive coding of speech signals and sub-
jective error criterion,” IEEE Transactions on Acoustics, Speech, and Signal
Processing, vol. ASSP-27, pp. 247–254, June 1979.
[7] B. S. Atal and M. R. Schroeder, “Improved quantizer for adaptive predictive
coding of speech signals at low bit rates,” Proceedings of ICASSP, pp. 535–538,
April 1980.
[8] T. B. Barnwell III, “Recursive windowing for generating autocorrelation anal-
ysis for LPC analysis,” IEEE Transactions on Acoustics, Speech and Signal
Processing, vol. ASSP-29, no. 5, pp. 1062–1066, October 1981.
[9] T. B. Barnwell III, K. Nayebi, and C. H. Richardson, Speech Coding, A Com-
puter Laboratory Textbook. John Wiley and Sons, 1996.
183
184 References
[10] L. E. Baum, “An inequality and associated maximization technique in statis-
tical estimation for probabilistic functions of Markov processes,” Inequalities,
vol. 3, pp. 1–8, 1972.
[11] L. E. Baum, T. Petri, G. Soules, and N. Weiss, “A maximization tech-
nique occurring in the statistical analysis of probabilistic functions of Markov
chains,” Annals in Mathematical Statistics, vol. 41, pp. 164–171, 1970.
[12] W. R. Bennett, “Spectra of quantized signals,” Bell System Technical Journal,
vol. 27, pp. 446–472, July 1948.
[13] M. Berouti, H. Garten, P. Kabal, and P. Mermelstein, “Efficient computation
and encoding of the multipulse excitation for LPC,” Proceedings of ICASSP,
pp. 384–387, March 1984.
[14] M. Beutnagel, A. Conkie, and A. K. Syrdal, “Diphone synthesis using
unit selection,” Third Speech Synthesis Workshop, Jenolan Caes, Australia,
November 1998.
[15] M. Beutnatel and A. Conkie, “Interaction of units in a unit selection
database,” Proceedings of Eurospeech ’99, Budapest, Hungary, September
1999.
[16] B. P. Bogert, M. J. R. Healy, and J. W. Tukey, “The quefrency alanysis of
times series for echos: Cepstrum, pseudo-autocovariance, cross-cepstrum, and
saphe cracking,” in Proceedings of the Symposium on Time Series Analysis,
(M. Rosenblatt, ed.), New York: John Wiley and Sons, Inc., 1963.
[17] E. Bresch, J. Nielsen, K. Nayak, and S. Narayanan, “Synchornized and noise-
robust audio recordings during realtime MRI scans,” Journal of the Acoustical
Society of America, vol. 120, no. 4, pp. 1791–1794, October 2006.
[18] C. S. Burrus and R. A. Gopinath, Introduction to Wavelets and Wavelet Trans-
forms. Prentice-Hall Inc., 1998.
[19] J. P. Campbell Jr., V. C. Welch, and T. E. Tremain, “An expandable error-
pretected 4800 bps CELP coder,” Proceedings of ICASSP, vol. 2, pp. 735–738,
May 1989.
[20] F. Charpentier and M. G. Stella, “Diphone synthesis using an overlap-add
technique for speech waveform concatenation,” Proceedings of International
Conference on Acoustics, Speech and Signal Processing, pp. 2015–2018, 1986.
[21] J. H. Chung and R. W. Schafer, “Performance evaluation of analysis-by-
synthesis homomorphic vocoders,” Proceedings of IEEE ICASSP, vol. 2,
pp. 117–120, March 1992.
[22] C. H. Coker, “A model of articulatory dynamics and control,” Proceedings of
IEEE, vol. 64, pp. 452–459, 1976.
[23] R. V. Cox, S. L. Gay, Y. Shoham, S. Quackenbush, N. Seshadri, and N. Jayant,
“New directions in subband coding,” IEEE Journal of Selected Areas in Com-
munications, vol. 6, no. 2, pp. 391–409, February 1988.
[24] R. E. Crochiere and L. R. Rabiner, Multirate Digital Signal Processing.
Prentice-Hall Inc., 1983.
[25] R. E. Crochiere, S. A. Webber, and J. L. Flanagan, “Digital coding of speech
in subbands,” Bell System Technical Journal, vol. 55, no. 8, pp. 1069–1085,
October 1976.
References 185
[26] C. C. Cutler, “Differential quantization of communication signals,” U.S.
Patent 2,605,361, July 29, 1952.
[27] S. B. Davis and P. Mermelstein, “Comparison of parametric representations
for monosyllabic word recognition in continuously spoken sentences,” IEEE
Transactions on Acoustics, Speech and Signal Processing, vol. 28, pp. 357–366,
August 1980.
[28] F. deJager, “Delta modulation — a new method of PCM transmission using
the 1-unit code,” Philips Research Reports, pp. 442–466, December 1952.
[29] P. B. Denes and E. N. Pinson, The speech chain. W. H. Freeman Company,
2nd Edition, 1993.
[30] H. Dudley, “The vocoder,” Bell Labs Record, vol. 17, pp. 122–126, 1939.
[31] T. Dutoit, An Introduction to Text-to-Speech Synthesis. Netherlands: Kluwer
Academic Publishers, 1997.
[32] G. Fant, Acoustic Theory of Speech Production. The Hague: Mouton & Co.,
1960; Walter de Gruyter, 1970.
[33] J. D. Ferguson, “Hidden Markov Analysis: An Introduction,” Hidden Markov
Models for Speech, Princeton: Institute for Defense Analyses, 1980.
[34] J. L. Flanagan, Speech Analysis, Synthesis and Perception. Springer-Verlag,
1972.
[35] J. L. Flanagan, C. H. Coker, L. R. Rabiner, R. W. Schafer, and N. Umeda,
“Synthetic voices for computers,” IEEE Spectrum, vol. 7, pp. 22–45, October
1970.
[36] J. L. Flanagan, K. Ishizaka, and K. L. Shipley, “Synthesis of speech from a
dynamic model of the vocal cords and vocal tract,” Bell System Technical
Journal, vol. 54, no. 3, pp. 485–506, March 1975.
[37] H. Fletcher and W. J. Munson, “Loudness, its definition, measurement and
calculation,” Journal of Acoustical Society of America, vol. 5, no. 2, pp. 82–
108, October 1933.
[38] G. D. Forney, “The Viterbi algorithm,” IEEE Proceedings, vol. 61, pp. 268–
278, March 1973.
[39] S. Furui, “Cepstral analysis technique for automatic speaker verification,”
IEEE Transactions on Acoustics Speech, and Signal Processing, vol. ASSP-
29, no. 2, pp. 254–272, April 1981.
[40] S. Furui, “Speaker independent isolated word recognition using dynamic fea-
tures of speech spectrum,” IEEE Transactions on Acoustics, Speech, Signal
Processing, vol. ASSP-26, no. 1, pp. 52–59, February 1986.
[41] O. Ghitza, “Audiotry nerve representation as a basis for speech processing,” in
Advances in Speech Signal Processing, (S. Furui and M. Sondhi, eds.), pp. 453–
485, NY: Marcel Dekker, 1991.
[42] J. J. Godfrey, E. C. Holliman, and J. McDaniel, “SWITCHBOARD: Telephone
Speech corpus for research and development,” Proceedings of ICASSP 1992,
pp. 517–520, 1992.
[43] B. Gold and L. R. Rabiner, “Parallel processing techniques for estimating
pitch period of speech in the time domain,” Journal of Acoustical Society of
America, vol. 46, no. 2, pt. 2, pp. 442–448, August 1969.
186 References
[44] A. L. Gorin, B. A. Parker, R. M. Sachs, and J. G. Wilpon, “How may I help
you?,” Proceedings of the Interactive Voice Technology for Telecommunications
Applications (IVTTA), pp. 57–60, 1996.
[45] R. M. Gray, “Vector quantization,” IEEE Signal Processing Magazine,
pp. 4–28, April 1984.
[46] R. M. Gray, “Toeplitz and circulant matrices: A review,” Foundations and
Trends in Communications and Information Theory, vol. 2, no. 3, pp. 155–239,
2006.
[47] J. A. Greefkes and K. Riemens, “Code modulation with digitally controlled
companding for speech transmission,” Philips Technical Review, pp. 335–353,
1970.
[48] H. Hermansky, “Auditory modeling in automatic recognition of speech,” in
Proceedings of First European Conference on Signal Analysis and Prediction,
pp. 17–21, Prague, Czech Republic, 1997.
[49] X. Huang, A. Acero, and H.-W. Hon, Spoken Language Processing. Prentice-
Hall Inc., 2001.
[50] A. Hunt and A. Black, “Unit selection in a concatenative speech synthesis
system using a large speech database,” Proceedings of ICASSP-96, Atlanta,
vol. 1, pp. 373–376, 1996.
[51] K. Ishizaka and J. L. Flanagan, “Synthesis of voiced sounds from a two-
mass model of the vocal cords,” Bell System Technical Journal, vol. 51, no. 6,
pp. 1233–1268, 1972.
[52] F. Itakura, “Line spectrum representation of linear predictive coefficients of
speech signals,” Journal of Acoustical Society of America, vol. 57, pp. 535(a),
p. s35(A).
[53] F. Itakura and S. Saito, “Analysis-synthesis telephony based upon the max-
imum likelihood method,” Proceedings of 6th International of Congress on
Acoustics, pp. C17–C20, 1968.
[54] F. Itakura and S. Saito, “A statistical method for estimation of speech spectral
density and formant frequencies,” Electronics and Communications in Japan,
vol. 53-A, no. 1, pp. 36–43, 1970.
[55] F. Itakura and T. Umezaki, “Distance measure for speech recognition based on
the smoothed group delay spectrum,” in Proceedings of ICASSP87, pp. 1257–
1260, Dallas TX, April 1987.
[56] N. S. Jayant, “Adaptive delta modulation with a one-bit memory,” Bell System
Technical Journal, pp. 321–342, March 1970.
[57] N. S. Jayant, “Adaptive quantization with one word memory,” Bell System
Technical Journal, pp. 1119–1144, September 1973.
[58] N. S. Jayant and P. Noll, Digital Coding of Waveforms. Prentice-Hall, 1984.
[59] F. Jelinek, Statistical Methods for Speech Recognition. Cambridge: MIT Press,
1997.
[60] F. Jelinek, R. L. Mercer, and S. Roucos, “Principles of lexical language model-
ing for speech recognition,” in Advances in Speech Signal Processing, (S. Furui
and M. M. Sondhi, eds.), pp. 651–699, Marcel Dekker, 1991.
References 187
[61] B. H. Juang, “Maximum likelihood estimation for mixture multivariate
stochastic observations of Markov chains,” AT&T Technology Journal, vol. 64,
no. 6, pp. 1235–1249, 1985.
[62] B. H. Juang, S. E. Levinson, and M. M. Sondhi, “Maximum likelihood esti-
mation for multivariate mixture observations of Markov chains,” IEEE Trans-
actions in Information Theory, vol. 32, no. 2, pp. 307–309, 1986.
[63] B. H. Juang, L. R. Rabiner, and J. G. Wilpon, “On the use of bandpass
liftering in speehc recognition,” IEEE Transactions on Acoustics, Speech and
Signal Processing, vol. ASSP-35, no. 7, pp. 947–954, July 1987.
[64] D. H. Klatt, “Software for a cascade/parallel formant synthesizer,” Journal of
the Acoustical Society of America, vol. 67, pp. 971–995, 1980.
[65] D. H. Klatt, “Review of text-to-speech conversion for English,” Journal of the
Acoustical Society of America, vol. 82, pp. 737–793, September 1987.
[66] W. Koenig, H. K. Dunn, and L. Y. Lacey, “The sound spectrograph,” Journal
of the Acoustical Society of America, vol. 18, pp. 19–49, 1946.
[67] P. Kroon, E. F. Deprettere, and R. J. Sluyter, “Regular-pulse excitation:
A nove approach to effective and efficient multipulse coding of speech,”
IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. ASSP-34,
pp. 1054–1063, October 1986.
[68] R. G. Leonard, “A database for speaker-independent digit recognition,” Pro-
ceedings of ICASSP 1984, pp. 42.11.1–42.11.4, 1984.
[69] S. E. Levinson, L. R. Rabiner, and M. M. Sondhi, “An introduction to the
application of the theory of probabilistic functions of a Markov process to
automatic speech recognition,” Bell System Technical Journal, vol. 62, no. 4,
pp. 1035–1074, 1983.
[70] Y. Linde, A. Buzo, and R. M. Gray, “An algorithm for vector quantizer
design,” IEEE Transactions on Communications, vol. COM-28, pp. 84–95,
January 1980.
[71] S. P. Lloyd, “Least square quantization in PCM,” IEEE Transactions on Infor-
mation Theory, vol. 28, pp. 129–137, March 1982.
[72] P. C. Loizou, Speech Enhancement, Theory and Practice. CRC Press, 2007.
[73] R. F. Lyon, “A computational model of filtering, detection and compression in
the cochlea,” in Proceedings of IEEE International Conference on Acoustics,
Speech and Signal Processing, Paris, France, May 1982.
[74] J. Makhoul, “Linear prediction: A tutorial review,” Proceedings of IEEE,
vol. 63, pp. 561–580, 1975.
[75] J. Makhoul, V. Viswanathan, R. Schwarz, and A. W. F. Huggins, “A mixed
source model for speech compression and synthesis,” Journal of the Acoustical
Society of America, vol. 64, pp. 1577–1581, December 1978.
[76] D. Malah, “Time-domain algorithms for harmonic bandwidth reduction and
time-scaling of pitch signals,” IEEE Transactions on Acoustics, Speech and
Signal Processing, vol. 27, no. 2, pp. 121–133, 1979.
[77] J. D. Markel, “The SIFT algorithm for fundamental frequency estimation,”
IEEE Transactions on Audio and Electroacoustics, vol. AU-20, no. 5, pp. 367–
377, December 1972.
[78] J. D. Markel and A. H. Gray, Linear Prediction of Speech. New York: Springer-
Verlag, 1976.
188 References
[79] J. Max, “Quantizing for minimum distortion,” IRE Transactions on Informa-
tion Theory, vol. IT-6, pp. 7–12, March 1960.
[80] A. V. McCree and T. P. Barnwell III, “A mixed excitation LPC vocoder
model for low bit rate speech coding,” IEEE Transactions on Speech and
Audio Processing, vol. 3, no. 4, pp. 242–250, July 1995.
[81] M. Mohri, “Finite-state transducers in language and speech processing,” Com-
putational Linguistics, vol. 23, no. 2, pp. 269–312, 1997.
[82] E. Moulines and F. Charpentier, “Pitch synchronous waveform processing
techniques for text-to-speech synthesis using diphones,” Speech Communica-
tion, vol. 9, no. 5–6, 1990.
[83] A. M. Noll, “Cepstrum pitch determination,” Journal of the Acoustical Society
of America, vol. 41, no. 2, pp. 293–309, February 1967.
[84] P. Noll, “A comparative study of various schemes for speech encoding,” Bell
System Technical Journal, vol. 54, no. 9, pp. 1597–1614, November 1975.
[85] A. V. Oppenehim, “Superposition in a class of nonlinear systems,” PhD dis-
sertation, MIT, 1964. Also: MIT Research Lab. of Electronics, Cambridge,
Massachusetts, Technical Report No. 432, 1965.
[86] A. V. Oppenheim, “A speech analysis-synthesis system based on homomorphic
filtering,” Journal of the Acoustical Society of America, vol. 45, no. 2, pp. 293–
309, February 1969.
[87] A. V. Oppenheim, “Speech spectrograms using the fast Fourier transform,”
IEEE Spectrum, vol. 7, pp. 57–62, August 1970.
[88] A. V. Oppenheim and R. W. Schafer, “Homomorphic analysis of speech,”
IEEE Transactions on Audio and Electroacoustics, vol. AU-16, pp. 221–228,
June 1968.
[89] A. V. Oppenheim, R. W. Schafer, and J. R. Buck, Discrete-Time Signal Pro-
cessing. Prentice-Hall Inc., 1999.
[90] A. V. Oppenheim, R. W. Schafer, and T. G. Stockham Jr., “Nonlinear filtering
of multiplied and convolved signals,” Proceedings of IEEE, vol. 56, no. 8,
pp. 1264–1291, August 1968.
[91] M. D. Paez and T. H. Glisson, “Minimum mean-squared error quantization in
speech,” IEEE Transactions on Communications, vol. Com-20, pp. 225–230,
April 1972.
[92] D. S. Pallett et al., “The 1994 benchmark tests for the ARPA spoken language
program,” Proceedings of 1995 ARPA Human Language Technology Work-
shop, pp. 5–36, 1995.
[93] M. R. Portnoff, “A quasi-one-dimensional simulation for the time-varying
vocal tract,” MS Thesis, MIT, Department of Electrical Engineering, 1973.
[94] T. F. Quatieri, Discrete-time speech signal processing. Prentice Hall, 2002.
[95] L. R. Rabiner, “A model for synthesizing speech by rule,” IEEE Transactions
on Audio and Electroacoustics, vol. AU-17, no. 1, pp. 7–13, March 1969.
[96] L. R. Rabiner, “On the use of autocorrelation analysis for pitch detection,”
IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. ASSP-25,
no. 1, pp. 24–33, February 1977.
[97] L. R. Rabiner, “A tutorial on hidden Markov models and selected applications
in speech recognition,” IEEE Proceedings, vol. 77, no. 2, pp. 257–286, 1989.
References 189
[98] L. R. Rabiner and B. H. Juang, “An introduction to hidden Markov models,”
IEEE Signal Processing Magazine, 1985.
[99] L. R. Rabiner and B. H. Juang, Fundamentals of Speech Recognition. Prentice-
Hall Inc., 1993.
[100] L. R. Rabiner and M. R. Sambur, “An algorithm for determining the endpoints
of isolated utterances,” Bell System Technical Journal, vol. 54, no. 2, pp. 297–
315, February 1975.
[101] L. R. Rabiner and R. W. Schafer, Theory and Application of Digital Speech
Processing. Prentice-Hall Inc., 2009. (In preparation).
[102] L. R. Rabiner, R. W. Schafer, and J. L. Flanagan, “Computer synthesis of
speech by concatenation of formant coded words,” Bell System Technical Jour-
nal, vol. 50, no. 5, pp. 1541–1558, May–June 1971.
[103] D. W. Robinson and R. S. Dadson, “A re-determination of the equal-loudness
contours for pure tones,” British Journal of Applied Physics, vol. 7, pp. 166–
181, 1956.
[104] R. C. Rose and T. P. Barnwell III, “The self excited vocoder — an alternate
approach to toll quality at 4800 bps,” Proceedings of ICASSP ’86, vol. 11,
pp. 453–456, April 1986.
[105] A. E. Rosenberg, “Effect of glottal pulse shape on the quality of natural vow-
els,” Journal of the Acoustical Society of America, vol. 43, no. 4, pp. 822–828,
February 1971.
[106] R. Rosenfeld, “Two decades of statistical language modeling: Where do we go
from here?,” IEEE Proceedings, vol. 88, no. 8, pp. 1270–1278, 2000.
[107] M. B. Sachs, C. C. Blackburn, and E. D. Young, “Rate-place and temporal-
place representations of vowels in the auditory nerve and anteroventral
cochlear nucleus,” Journal of Phonetics, vol. 16, pp. 37–53, 1988.
[108] Y. Sagisaka, “Speech synthesis by rule using an optimal selection of non-
uniform synthesis units,” Proceedings of the International Conference on
Acoustics, Speech and Signal Processing, pp. 679–682, 1988.
[109] R. W. Schafer, “Echo removal by discrete generalized linear filtering,” PhD
dissertation, MIT, 1968. Also: MIT Research Laboratory of Electronics, Cam-
bridge, Massachusetts, Technical Report No. 466, 1969.
[110] R. W. Schafer, “Homomorphic systems and cepstrum analysis of speech,”
Springer Handbook of Speech Processing and Communication, Springer, 2007.
[111] R. W. Schafer and L. R. Rabiner, “System for automatic formant analysis of
voiced speech,” Journal of the Acoustical Society of America, vol. 47, no. 2,
pp. 458–465, February 1970.
[112] M. R. Schroeder and B. S. Atal, “Code-excited linear prediction (CELP):
High-quality speech at very low bit rates,” Proceedings of IEEE ICASSP,
pp. 937–940, 1985.
[113] M. R. Schroeder and E. E. David, “A vocoder for transmitting 10 kc/s speech
over a 3.5 kc/s channel,” Acustica, vol. 10, pp. 35–43, 1960.
[114] J. H. Schroeter, “Basic principles of speech synthesis,” Springer Handbook of
Speech Processing, Springer-Verlag, 2006.
[115] S. Seneff, “A joint synchrony/mean-rate model of auditory speech processing,”
Journal of Phonetics, vol. 16, pp. 55–76, 1988.
190 References
[116] C. E. Shannon and W. Weaver, The Mathematical Theory of Communication.
University of Illinois Press, Urbana, 1949.
[117] G. A. Sitton, C. S. Burrus, J. W. Fox, and S. Treitel, “Factoring very-high-
degree polynomials,” IEEE Signal Processing Magazine, vol. 20, no. 6, pp. 27–
42, November 2003.
[118] B. Smith, “Instantaneous companding of quantized signals,” Bell System Tech-
nical Journal, vol. 36, no. 3, pp. 653–709, May 1957.
[119] F. K. Soong and B.-H. Juang, “Optimal quantization of LSP parameters,”
IEEE Transactions on Speech and Audio Processing, vol. 1, no. 1, pp. 15–24,
January 1993.
[120] A. Spanias, T. Painter, and V. Atti, Audio Signal Processing and Coding.
Wiley Interscience, 2007.
[121] K. N. Stevens, Acoustic Phonetics. MIT Press, 1998.
[122] S. S. Stevens and J. Volkman, “The relation of pitch to frequency,” American
Journal of Psychology, vol. 53, p. 329, 1940.
[123] L. C. Stewart, R. M. Gray, and Y. Linde, “The design of trellis waveform
coders,” IEEE transactions on Communications, vol. COM-30, pp. 702–710,
April 1982.
[124] T. G. Stockham Jr., T. M. Cannon, and R. B. Ingebretsen, “Blind deconvolu-
tion through digital signal processing,” Proceedings of IEEE, vol. 63, pp. 678–
692, April 1975.
[125] G. Strang and T. Nguyen, Wavelets and Filter Banks. Wellesley, MA:
Wellesley-Cambridge Press, 1996.
[126] Y. Tohkura, “A weighted cepstral distance measure for speech recognition,”
IEEE Transactions on Acoustics, Speech and Signal Processing, vol. 35,
pp. 1414–1422, October 1987.
[127] J. M. Tribolet, “A new phase unwrapping algorithm,” IEEE Transactions on
Acoustical, Speech, and Signal Processing, vol. ASSP-25, no. 2, pp. 170–177,
April 1977.
[128] C. K. Un and D. T. Magill, “The residual-excited linear prediction vocoder
with transmission rate below 9.6 kbits/s,” IEEE Transactions on Communi-
cations, vol. COM-23, no. 12, pp. 1466–1474, December 1975.
[129] P. P. Vaidyanathan, Multirate Systems and Filter Banks. Prentice-Hall Inc.,
1993.
[130] R. Viswanathan, W. Russell, and J. Makhoul, “Voice-excited LPC coders for
9.6 kbps speech transmission,” vol. 4, pp. 558–561, April 1979.
[131] V. Viswanathan and J. Makhoul, “Quantization properties of transmission
parameters in linear predictive systems,” IEEE Transactions on Acoustics,
Speech, and Signal Processing, vol. ASSP-23, no. 3, pp. 309–321, June 1975.
[132] A. J. Viterbi, “Error bounds for convolutional codes and an asymptotically
optimal decoding Aalgorithm,” IEEE Transactions on Information Theory,
vol. IT-13, pp. 260–269, April 1967.
[133] W. Ward, “Evaluation of the CMU ATIS system,” Proceedings of DARPA
Speech and Natural Language Workshop, pp. 101–105, February 1991.
[134] E. Zwicker and H. Fastl, Psycho-acoustics. Springer-Verlag, 2nd Edition, 1990.
Supplemental References
The specific references that comprise the Bibliography of this text
are representative of the literature of the field of digital speech pro-
cessing. In addition, we provide the following list of journals and
books as a guide for further study. Listing the books in chronologi-
cal order of publication provides some perspective on the evolution of
the field.
Speech Processing Journals
• IEEE Transactions on Signal Processing. Main publication of IEEE Sig-
nal Processing Society.
• IEEE Transactions on Speech and Audio. Publication of IEEE Signal
Processing Society that is focused on speech and audio processing.
• Journal of the Acoustical Society of America. General publication of the
American Institute of Physics. Papers on speech and hearing as well as
other areas of acoustics.
• Speech Communication. Published by Elsevier. A publication of the
European Association for Signal Processing (EURASIP) and of the
International Speech Communication Association (ISCA).
191
192 Supplemental References
General Speech Processing References
• Speech Analysis, Synthesis and Perception, J. L. Flanagan, Springer-
Verlag, Second Edition, Berlin, 1972.
• Linear Prediction of Speech, J. D. Markel and A. H. Gray, Jr., Springer
Verlag, Berlin, 1976.
• Digital Processing of Speech Signals, L. R. Rabiner and R. W. Schafer,
Prentice-Hall Inc., 1978.
• Speech Analysis, R. W. Schafer and J. D. Markel (eds.), IEEE Press
Selected Reprint Series, 1979.
• Speech Communication, Human and Machine, D. O’Shaughnessy,
Addison-Wesley, 1987.
• Advances in Speech Signal Processing, S. Furui and M. M. Sondhi, Mar-
cel Dekker Inc., New York, 1991.
• Discrete-Time Processing of Speech Signals, J. Deller, Jr., J. H. L.
Hansen, and J. G. Proakis, Wiley-IEEE Press, Classic Reissue, 1999.
• Acoustic Phonetics, K. N. Stevens, MIT Press, 1998.
• Speech and Audio Signal Processing, B. Gold and N. Morgan, John Wiley
and Sons, 2000.
• Digital Speech Processing, Synthesis and Recognition, S. Furui, Second
Edition, Marcel Dekker Inc., New York, 2001.
• Discrete-Time Speech Signal Processing, T. F. Quatieri, Prentice Hall
Inc., 2002.
• Speech Processing, A Dynamic and Optimization-Oriented Approach,
L. Deng and D. O’Shaughnessy, Marcel Dekker, 2003.
• Springer Handbook of Speech Processing and Speech Communication,
J. Benesty, M. M. Sondhi and Y Huang (eds.), Springer, 2008.
• Theory and Application of Digital Speech Processing, L. R. Rabiner and
R. W. Schafer, Prentice Hall Inc., 2009.
Speech Coding References
• Digital Coding of Waveforms, N. S. Jayant and P. Noll, Prentice Hall
Inc., 1984.
• Practical Approaches to Speech Coding, P. E. Papamichalis, Prentice
Hall Inc., 1987.
• Vector Quantization and Signal Compression, A. Gersho and R. M.
Gray, Kluwer Academic Publishers, 1992.
• Speech Coding and Synthesis, W. B. Kleijn and K. K. Paliwal, Elsevier,
1995.
• Speech Coding, A Computer Laboratory Textbook, T. P. Barnwell and
K. Nayebi, John Wiley and Sons, 1996.
Supplemental References 193
• A Practical Handbook of Speech Coders, R. Goldberg and L. Riek, CRC
Press, 2000.
• Speech Coding Algorithms, W. C. Chu, John Wiley and Sons, 2003.
• Digital Speech: Coding for Low Bit Rate Communication Systems, Sec-
ond Edition, A. M. Kondoz, John Wiley and Sons, 2004.
Speech Synthesis
• From Text to Speech, J. Allen, S. Hunnicutt and D. Klatt, Cambridge
University Press, 1987.
• Acoustics of American English, J. P. Olive, A. Greenwood and J. Cole-
man, Springer-Verlag, 1993.
• Computing Prosody, Y. Sagisaka, N. Campbell and N. Higuchi, Springer-
Verlag, 1996.
• Progress in Speech Synthesis, J. VanSanten, R. W. Sproat, J. P. Olive
and J. Hirschberg (eds.), Springer-Verlag, 1996.
• An Introduction to Text-to-Speech Synthesis, T. Dutoit, Kluwer Aca-
demic Publishers, 1997.
• Speech Processing and Synthesis Toolboxes, D. Childers, John Wiley and
Sons, 1999.
• Text To Speech Synthesis: New Paradigms and Advances, S. Narayanan
and A. Alwan (eds.), Prentice Hall Inc., 2004.
• Text-to-Speech Synthesis, P. Taylor, Cambridge University Press, 2008.
Speech Recognition and Natural Language Processing
• Fundamentals of Speech Recognition, L. R. Rabiner and B. H. Juang,
Prentice Hall Inc., 1993.
• Connectionist Speech Recognition-A Hybrid Approach, H. A. Bourlard
and N. Morgan, Kluwer Academic Publishers, 1994.
• Automatic Speech and Speaker Recognition, C. H. Lee, F. K. Soong and
K. K. Paliwal (eds.), Kluwer Academic Publisher, 1996.
• Statistical Methods for Speech Recognition, F. Jelinek, MIT Press, 1998.
• Foundations of Statistical Natural Language Processing, C. D. Manning
and H. Schutze, MIT Press, 1999.
• Spoken Language Processing, X. Huang, A. Acero and H.-W. Hon, Pren-
tice Hall Inc., 2000.
• Speech and Language Processing, D. Jurafsky and J. H. Martin, Prentice
Hall Inc., 2000.
• Mathematical Models for Speech Technology, S. E. Levinson, John Wiley
and Sons, 2005.
194 Supplemental References
Speech Enhancement
• Digital Speech Transmission, Enhancement, Coding and Error Conceal-
ment, P. Vary and R. Martin, John Wiley and Sons, Ltd., 2006.
• Speech Enhancement, Theory and Practice, P. C. Loizou, CRC Press,
2007.
Audio Processing
• Applications of Digital Signal Processing to Audio and Acoustics,
H. Kahrs and K. Brandenburg (eds.), Kluwer Academic Publishers,
1998.
• Audio Signal Processing and Coding, A. Spanias, T. Painter and V. Atti,
John Wiley and Sons, 2007.

aspect of speech processing to great depth; hence our goal is to provide a useful introduction to the wide range of important concepts that comprise the field of digital speech processing. A more comprehensive treatment will appear in the forthcoming book, Theory and Application of Digital Speech Processing [101].

1
Introduction

The fundamental purpose of speech is communication, i.e., the transmission of messages. According to Shannon’s information theory [116], a message represented as a sequence of discrete symbols can be quantified by its information content in bits, and the rate of transmission of information is measured in bits/second (bps). In speech production, as well as in many human-engineered electronic communication systems, the information to be transmitted is encoded in the form of a continuously varying (analog) waveform that can be transmitted, recorded, manipulated, and ultimately decoded by a human listener. In the case of speech, the fundamental analog form of the message is an acoustic waveform, which we call the speech signal. Speech signals, as illustrated in Figure 1.1, can be converted to an electrical waveform by a microphone, further manipulated by both analog and digital signal processing, and then converted back to acoustic form by a loudspeaker, a telephone handset or headphone, as desired. This form of speech processing is, of course, the basis for Bell’s telephone invention as well as today’s multitude of devices for recording, transmitting, and manipulating speech and audio signals. Although Bell made his invention without knowing the fundamentals of information theory, these ideas
3

02 0. 1. In their classic introduction to speech science. it is nevertheless useful to begin with a discussion of how information is encoded in the speech waveform. and finally to the understanding of the message by a listener.” have assumed great importance in the design of sophisticated modern communications systems.12 W IY CH 0.04 0. 1.06 time in seconds 0.2). The process starts in the upper left as a message represented somehow in the brain of the speaker. The message information can be thought of as having a number of different representations during the process of speech production (the upper path in Figure 1.4 Introduction SH 0 UH D 0.2 shows the complete process of producing and perceiving speech from the formulation of a message in the brain of a talker.24 EY 0.36 S 0. . to the creation of the speech signal. Therefore.1 A speech waveform with phonetic labels for the text message “Should we chase. Denes and Pinson aptly referred to this process as the “speech chain” [29].1 0.12 Fig.48 0 0.1 The Speech Chain Figure 1. even though our main focus will be mostly on the speech waveform and its representation in the form of parametric models.08 0.

2. prosody Language Code discrete input articulatory motions Neuro-Muscular Controls continuous input excitation. the segments of the waveform of Figure 1. For example the message could be represented initially as English text. teeth.1 The Speech Chain Speech Production text Message Formulation phonemes. The ARPAbet code does not require special fonts and is thus more convenient for computer applications. the text “should we chase” is represented phonetically (in ARPAbet symbols) as [SH UH D — W IY — CH EY S].” i. 1. to speech signal. namely the tongue. This step. the set of control signals that direct the neuro-muscular system to move the speech articulators. to understanding.) The third step in the speech production process is the conversion to “neuro-muscular controls. the speed and emphasis) in which the sounds are intended to be produced.e. (See Chapter 2 for more discussion of phonetic transcription.1. 1 The International Phonetic Association (IPA) provides a set of rules for phonetic transcription using an equivalent set of specialized symbols. the talker implicitly converts the text into a symbolic representation of the sequence of sounds corresponding to the spoken version of the text. words sentences Language Translation discrete output spectrum analysis Basilar Neural Membrane Transduction Motion continuous output feature extraction acoustic waveform Speech Perception Fig. lips.2 The Speech Chain: from message.. In order to “speak” the message. converts text symbols to phonetic symbols (along with stress and durational information) that describe the basic sounds of a spoken version of the message and the manner (i.e. called the language code generator in Figure 1. . formants Vocal Tract System acoustic waveform 5 50 bps 200 bps 2000 bps information rate 64-700 Kbps Transmission Channel semantics Message Understanding phonemes.1 are labeled with phonetic symbols using a computer-keyboard-friendly code called ARPAbet.1 Thus. As an example..

Furthermore. At the second stage of the process.g.1 contains approximately 64 = 26 symbols. assume that there are about 32 symbols (letters) in the language (in English there are 26 letters. The information representations for the first two stages in the speech chain are discrete so we can readily estimate the rate of information flow with some simple assumptions. If they could be measured.6 = 80 bps. The end result of the neuro-muscular controls step is a set of articulatory motions (continuous control) that cause the vocal tract articulators to move in a prescribed manner in order to create the desired sounds.g. or about 6 bits/phoneme (again a rough approximation assuming independence of phonemes). where the text representation is converted into phonemes and prosody (e.1. but still acceptable for a rough information rate estimate).1. in a manner that is consistent with the sounds of the desired spoken message and with the desired degree of emphasis.6 Introduction jaw and velum. the rate of speaking for most people is about 10 symbols per second (somewhat on the high side. duration. pitch and stress) markers. For example.. This leads to an estimate of 8 × 6/0. pitch. the information rate is estimated to increase by a factor of 4 to about 200 bps. For the next stage in the speech production part of the speech chain. there are 8 phonemes in approximately 600 ms. that encodes the information in the desired message into the speech signal. Finally the last step in the Speech Production process is the “vocal tract system” that physically creates the necessary sound sources and the appropriate vocal tract shapes over time so as to create an acoustic waveform. loudness) could easily add 100 bps to the total information rate for a message encoded as a speech signal. we could estimate the spectral bandwidth of these . the ARBAbet phonetic symbol set used to label the speech sounds in Figure 1. such as the one shown in Figure 1. but if we include simple punctuation we get a count closer to 32 = 25 symbols). Hence. To determine the rate of information flow during speech production. we estimate the base information rate of the text message as about 50 bps (5 bits per symbol times 10 symbols per second). the representation becomes continuous (in the form of control signals for articulatory motion).. Additional information required to describe prosodic features of the signal (e. assuming independent letters as a simple approximation. In Figure 1.

For example. The above analysis of data rates shows that as we move from text to sampled speech waveform. .000 to more than 700. “telephone quality” requires that a bandwidth of 0–4 kHz be preserved.000 bps. it will sound different from the original speech signal uttered by the talker.000 bps.600 bps. This representation is highly intelligible (i. the reproduced acoustic signal will be virtually indistinguishable from the original speech signal. or a data rate of 705. Each sample can be quantized with 8 bits on a log scale.1. the result is an encoding of the message that can be effectively transmitted by acoustic wave propagation and robustly decoded by the hearing mechanism of a listener. On the other hand..100 samples/s with 16 bit samples.e. the data rate of the digitized speech waveform at the end of the speech production part of the speech chain can be anywhere from 64. the data rate can increase by a factor of 10. Thus. etc. but much of it is due to the inefficiency 2 Note that we introduce the term data rate for digital representations to distinguish from the inherent information content of the message represented by the speech signal. resulting in a bit rate of 64.2 Finally. as we will see later. Part of this extra information represents characteristics of the talker such as emotional state.000.. accent. As we move from text to speech waveform through the speech chain. humans can readily extract the message from it) but to most listeners. Estimates of bandwidth and required accuracy suggest that the total data rate of the sampled articulatory control signals is about 2000 bps [34]. The articulators move relatively slowly compared to the time variation of the resulting acoustic waveform.1 The Speech Chain 7 control signals and appropriately sample and quantize these signals to obtain equivalent digital signals for which the data rate could be estimated. speech mannerisms. implying a sampling rate of 8000 samples/s. In this case. the speech waveform can be represented with “CD quality” using a sampling rate of 44. We arrive at such numbers by examining the sampling rate and quantization required to represent the speech signal with a desired perceptual fidelity. the original text message is represented by a set of continuously varying signals whose digital representation requires a much higher data rate than the information rate that we estimated for transmission of the message as a speech signal.

and thus the entire model is useful for thinking about the processes that occur.2. Thus.8 Introduction of simply sampling and finely quantizing analog signals. and sentences associated with the in-coming message by a language translation process in the human brain. words. but it is generally agreed that some physical correlate of each of the steps in the speech perception model occur within the human brain. Finally. In its simplest embodiment. this transmission channel consists of just the acoustic wave connection between . the last step in the speech perception model is the conversion of the phonemes. as well as a speech perception/recognition model. which acts as a non-uniform spectrum analyzer by spatially separating the spectral components of the incoming speech signal and thereby analyzing them by what amounts to a non-uniform filter bank. The next step in the process is a conversion of the sound features into the set of phonemes. The speech perception model shows the series of steps from capturing speech at the ear to understanding the message encoded in the speech signal.2 that we have not discussed — namely the transmission channel between the speech generation and speech perception parts of the model. of the type discussed above. There is one additional process shown in the diagram of the complete speech chain in Figure 1. The first step is the effective conversion of the acoustic waveform to a spectral representation. The next step in the speech perception process is a neural transduction of the spectral features into a set of sound features (or distinctive features as they are referred to in the area of linguistics) that can be decoded and processed by the brain. as shown progressing to the left in the bottom half of Figure 1. words and sentences of the message into an understanding of the meaning of the basic message in order to be able to respond to or take some appropriate action. This is done within the inner ear by the basilar membrane.2 is rudimentary at best. Our fundamental understanding of the processes in most of the speech perception modules in Figure 1. The complete speech chain consists of a speech production/ generation model. a central theme of much of digital speech processing is to obtain a digital representation with lower data rate than that of the sampled waveform. motivated by an awareness of the low intrinsic information rate of speech.

1.2 Applications of Digital Speech Processing

9

a speaker and a listener who are in a common space. It is essential to include this transmission channel in our model for the speech chain since it includes real world noise and channel distortions that make speech and message understanding more difficult in real communication environments. More interestingly for our purpose here — it is in this domain that we find the applications of digital speech processing.

1.2

Applications of Digital Speech Processing

The first step in most applications of digital speech processing is to convert the acoustic waveform to a sequence of numbers. Most modern A-to-D converters operate by sampling at a very high rate, applying a digital lowpass filter with cutoff set to preserve a prescribed bandwidth, and then reducing the sampling rate to the desired sampling rate, which can be as low as twice the cutoff frequency of the sharp-cutoff digital filter. This discrete-time representation is the starting point for most applications. From this point, other representations are obtained by digital processing. For the most part, these alternative representations are based on incorporating knowledge about the workings of the speech chain as depicted in Figure 1.2. As we will see, it is possible to incorporate aspects of both the speech production and speech perception process into the digital representation and processing. It is not an oversimplification to assert that digital speech processing is grounded in a set of techniques that have the goal of pushing the data rate of the speech representation to the left along either the upper or lower path in Figure 1.2. The remainder of this chapter is devoted to a brief summary of the applications of digital speech processing, i.e., the systems that people interact with daily. Our discussion will confirm the importance of the digital representation in all application areas.

1.2.1

Speech Coding

Perhaps the most widespread applications of digital speech processing technology occur in the areas of digital transmission and storage

10 Introduction
speech signal data

xc (t )

A-to-D Converter

samples

x[n]

Analysis/ Encoding

y[n] Channel or Medium

decoded signal

xc (t )

D-to-A Converter

samples

x[n]

Synthesis/ Decoding

data

y[n]

Fig. 1.3 Speech coding block diagram — encoder and decoder.

of speech signals. In these areas the centrality of the digital representation is obvious, since the goal is to compress the digital waveform representation of speech into a lower bit-rate representation. It is common to refer to this activity as “speech coding” or “speech compression.” Figure 1.3 shows a block diagram of a generic speech encoding/decoding (or compression) system. In the upper part of the figure, the A-to-D converter converts the analog speech signal xc (t) to a sampled waveform representation x[n]. The digital signal x[n] is analyzed and coded by digital computation algorithms to produce a new digital signal y[n] that can be transmitted over a digital communication channel or stored in a digital storage medium as y [n]. As we will see, there ˆ are a myriad of ways to do the encoding so as to reduce the data rate over that of the sampled and quantized speech waveform x[n]. Because the digital representation at this point is often not directly related to the sampled speech waveform, y[n] and y [n] are appropriately referred ˆ to as data signals that represent the speech signal. The lower path in Figure 1.3 shows the decoder associated with the speech coder. The received data signal y [n] is decoded using the inverse of the analysis ˆ processing, giving the sequence of samples x[n] which is then converted ˆ (using a D-to-A Converter) back to an analog signal xc (t) for human ˆ listening. The decoder is often called a synthesizer because it must reconstitute the speech waveform from data that may bear no direct relationship to the waveform.

1.2 Applications of Digital Speech Processing

11

With carefully designed error protection coding of the digital representation, the transmitted (y[n]) and received (ˆ[n]) data can be y essentially identical. This is the quintessential feature of digital coding. In theory, perfect transmission of the coded digital representation is possible even under very noisy channel conditions, and in the case of digital storage, it is possible to store a perfect copy of the digital representation in perpetuity if sufficient care is taken to update the storage medium as storage technology advances. This means that the speech signal can be reconstructed to within the accuracy of the original coding for as long as the digital representation is retained. In either case, the goal of the speech coder is to start with samples of the speech signal and reduce (compress) the data rate required to represent the speech signal while maintaining a desired perceptual fidelity. The compressed representation can be more efficiently transmitted or stored, or the bits saved can be devoted to error protection. Speech coders enable a broad range of applications including narrowband and broadband wired telephony, cellular communications, voice over internet protocol (VoIP) (which utilizes the internet as a real-time communications medium), secure voice for privacy and encryption (for national security applications), extremely narrowband communications channels (such as battlefield applications using high frequency (HF) radio), and for storage of speech for telephone answering machines, interactive voice response (IVR) systems, and prerecorded messages. Speech coders often utilize many aspects of both the speech production and speech perception processes, and hence may not be useful for more general audio signals such as music. Coders that are based on incorporating only aspects of sound perception generally do not achieve as much compression as those based on speech production, but they are more general and can be used for all types of audio signals. These coders are widely deployed in MP3 and AAC players and for audio in digital television systems [120]. 1.2.2 Text-to-Speech Synthesis

For many years, scientists and engineers have studied the speech production process with the goal of building a system that can start with

The first block in the text-to-speech synthesis system. Instead. There are many procedures for assembling the speech sounds and compiling them into a proper sentence. how to pronounce ambiguous words like read. etc. text and produce speech automatically. the role of the synthesis algorithm is to create the appropriate sound sequence to represent the text message in the form of speech. 1. half phones. and how to properly pronounce proper names. a text-to-speech synthesizer such as depicted in Figure 1. labeled linguistic rules. The basic digital representation is not generally the sampled speech wave. has the job of converting the printed text input into a set of sounds that the machine must synthesize. the synthesis algorithm must simulate the action of the vocal tract system in creating the sounds of speech. This is more difficult than simply looking up the words in a pronouncing dictionary because the linguistic rules must determine how to pronounce acronyms. rates of speaking. (street or Saint). specialized terms. how to pronounce abbreviations like St. and then decides which sequence of speech units sounds best for the particular text message that is being produced. bass. object. Dr. In a sense.) so that the resulting synthetic speech will express the words and intent of the text message in what passes for a natural voice that can be decoded accurately by human speech perception. the computer stores multiple versions of each of the basic units of speech (phones. but the most promising one today is called “unit selection and concatenation.4 is a digital simulation of the entire upper part of the speech chain diagram. (Doctor or drive).” In this method. some sort of compressed representation is normally used to . etc. The conversion from text to sounds involves a set of linguistic rules that must determine the appropriate set of sounds (perhaps including things like emphasis. Once the proper pronunciation of the text has been determined.).4 Text-to-speech synthesis system block diagram.12 Introduction text Linguistic Rules Synthesis Algorithm D-to-A Converter speech Fig. etc. The input to the system is ordinary text such as an email message or an article from a newspaper or magazine. In essence. syllables. pauses.

more importantly. the speech synthesis algorithm would include an appropriate decoder. dictionaries. provide the voices for talking agents for completion of transactions over the internet.1. where the object is to extract the message from the speech signal.5 Block diagram of general pattern matching system for speech signals. Most such systems involve some sort of pattern matching. 1. 1. updates on arrivals and departures of flights. speaker recognition.2.2 Applications of Digital Speech Processing 13 save memory and. Text-to-speech synthesis systems are an essential component of modern human–machine communications systems and are used to do things like read email messages over a telephone. and as the voice of announcement machines that provide information such as stock quotes. Thus. airline schedules. serve as the voice for providing information from handheld devices such as foreign language phrasebooks. where an optical character recognition system provides the text input to a speech synthesis system. where the goal is to verify a speaker’s claimed identity from analysis of their speech speech A-to-D Converter Feature Analysis Pattern Matching symbols Fig.2.5 shows a block diagram of a generic approach to pattern matching problems in speech processing. crossword puzzle helpers. speaker verification. to allow convenient manipulation of durations and blending of adjacent sounds.1. provide voice output from GPS systems in automobiles. . Figure 1. etc. as discussed in Section 1.3 Speech Recognition and Other Pattern Matching Problems Another large class of digital speech processing applications is concerned with the automatic extraction of information from the speech signal. handle call center help desks and customer care applications. Another important application is in reading machines for the blind. Such problems include the following: speech recognition. where the goal is to identify who is speaking. whose output is converted to an analog representation via the D-to-A converter.

The symbolic output consists of a set of recognized words. which involves monitoring a speech signal for the occurrence of specified words or phrases. namely the pattern matching block. the biggest use has been in the area of recognition and understanding of speech in support of human– machine communication by voice.14 Introduction signal.5 represents a wide range of speech pattern matching problems. the same analysis techniques that are used in speech coding are also used to derive the feature vectors. etc. word spotting. Often. and chooses the identity associated with the pattern which is the closest match to the time-aligned set of feature vectors of the speech signal. The final block in the system. and other documents. address list modification and entry. voice dictation to create letters. Names from directories with upwards of several hundred names can readily be recognized and dialed using simple speech recognition technology. and automatic indexing of speech recordings based on recognition (or spotting) of spoken keywords. dynamically time aligns the set of feature vectors representing the speech signal with a concatenated set of stored patterns. in the case of speaker recognition. natural language voice dialogues with machines to enable help desks and call centers. Spoken name speech recognition in cellphones enables voice dialing capability that can automatically dial the number associated with the recognized name. and for agent services such as calendar entry and update. memos. . or the identity of the best matching talker. in the case of speech recognition. For example. The first block in the pattern matching system converts the analog speech waveform to digital form using an A-to-D converter. Speech coding at bit rates on the order of 8 Kbps enables normal voice conversations in cell phones. or a decision as to whether to accept or reject the identity claim of a speaker in the case of speaker verification. one of the preeminent uses of speech technology is in portable communication devices. Although the block diagram of Figure 1. The major areas where such a system finds applications include command and control of computer software. Pattern recognition applications often occur in conjunction with other digital speech processing applications. The feature analysis module converts the sampled speech signal to a set of feature vectors.

it can be manipulated in virtually limitless ways by DSP techniques. speech signal quality enhancement. and aids for the hearing. The goal of language translation systems is to convert spoken words in one language to spoken words in another language so as to facilitate natural language voice dialogues between people speaking different languages. the techniques of digital speech processing are a key ingredient of a wide range of applications that include the three areas of transmission/storage. and speech recognition as well as many others such as speaker identification. . along with speech recognition (and generally natural language understanding) that also works for both languages. Language translation technology requires speech synthesis systems that work in both languages. hence it is a very difficult task and one for which only limited progress has been made.2 Applications of Digital Speech Processing 15 Another major speech application that has long been a dream of speech researchers is automatic language translation.2.4 Other Speech Applications The range of speech communication applications is illustrated in Figure 1. it will be possible for people speaking different languages to communicate at data rates on the order of that of printed text reading! 1. When such systems exist. Here again.1.7 represents any system where time signals such as speech are processed by the techniques of DSP. 1.6. manipulations and modifications of the speech signal are Digital Speech Processing Techniques Digital Speech Transmission Synthesis & Storage Speech Recognition Speaker Enhancement Aids for the Verification/ of Speech Handicapped Identification Quality Fig. speech synthesis.or visuallyimpaired. As seen in this figure. This figure simply depicts the notion that once the speech signal is sampled.6 Range of speech communication applications. The block diagram in Figure 1.

to modify voice qualities. 1. where the goal is to remove or suppress noise or echo or reverberation picked up by a microphone along with the desired speech signal. An excellent reference in this area is the recent textbook by Loizou [72]. however.3 Our Goal for this Text We have discussed the speech signal and how it encodes information for human communication. in making distorted speech signals more useful for further processing as part of a speech coder. synthesizer. using a D-to-A converter.16 Introduction speech A-to-D Converter Computer Algorithm D-to-A Converter speech Fig. and to speed-up or slow-down prerecorded speech (e. The subject is too broad and .7 General block diagram for application of digital signal processing to speech signals.. in reality the best that has been achieved so far is less perceptually annoying speech that essentially maintains. however. but does not improve. One important application area is speech enhancement. We make no pretense of exhaustive coverage. or careful scrutinizing of spoken material). Success has been achieved. These and many more examples all rely on the basic principles of digital speech processing. operating on that representation by further digital computation. the intelligibility of the noisy speech.g. 1. rapid review of voice mail messages. for talking books. and then transforming back to the waveform domain. usually achieved by transforming the speech signal into an alternative representation (that is motivated by our understanding of speech production and speech perception). or recognizer. In human-to-human communication. and we have hinted at some of the possibilities that exist for the future. Other examples of manipulation of the speech signal include timescale modification to align voices with video segments. which we will discuss in the remainder of this text. the goal of speech enhancement systems is to make the speech more intelligible and more natural. We have given a brief overview of the way in which digital speech processing is being applied today.

Our goal is only to provide an up-to-date introduction to this fascinating field.3 Our Goal for this Text 17 too deep. This means that some of the latest algorithmic innovations and applications will not be discussed — not because they are not interesting. We hope that this text will stimulate readers to investigate the subject in greater depth using the extensive set of references provided. and recognition.1. Instead our focus is on the fundamentals of digital speech processing and their application to coding. . but simply because there are so many fundamental tried-and-true techniques that remain at the core of digital speech processing. synthesis. and we will not be able to cover all the possible applications of digital speech processing techniques. We will not be able to go into great depth.

A condensed inventory of the sounds of speech in the English language is given in Table 2. For most languages the number of phonemes is between 32 and 64.1 also includes some simple examples of ARPAbet transcriptions of words containing each of the phonemes of English. Table 2. the number of which depends upon the language and the refinement of the analysis.1. this chapter provides a brief overview of the phonetic representation of speech and an introduction to models for the production of the speech signal. 2. 18 . where the phonemes are denoted by a set of ASCII symbols called the ARPAbet. Additional phonemes can be added to Table 2.2 The Speech Signal As the discussion in Chapter 1 shows.1 Phonetic Representation of Speech Speech can be represented phonetically by a finite set of symbols called the phonemes of the language.1 to account for allophonic variations and events such as glottal stops and pauses. To gain a better idea of what this means. the goal in many applications of digital speech processing techniques is to move the digital representation of the speech signal from the waveform samples back up the speech chain toward the message.

edu/cgi-bin/cmudict.cmu.1 on p. Thus sounds like /SH/ and /S/ look like (spectrally shaped) . Figure 1.2. for the most part.speech. Class Vowels and diphthongs ARPAbet IY IH EY EH AE AA AO UH OW UW AH ER AY AW OY Y R W L M N NG P B T D K G HH F V TH DH S Z SH ZH CH JH Example beet bit bait bet bat bob born book boat boot but bir d buy dow n boy you r ent w it let met net sing pat bet ten debt k it get hat f at v at thing that sat z oo shut az ure chase j udge Transcription [B IY T] [B IH T] [B EY T] [B EH T] [B AE T] [B AA B] [B AO R N] [B UH K] [B OW T] [B UW T] [B AH T ] [B ER D] [B AY] [D AW N] [B OY] [Y UH] [R EH N T] [W IH T] [L EH T] [M EH T] [N EH T] [S IH NG] [P AE T] [B EH T] [T EH N] [D EH T] [K IH T] [G EH T] [HH AE T] [F AE T] [V AE T] [TH IH NG] [DH AE T] [S AE T] [Z UW] [SH AH T] [AE ZH ER] [CH EY S] [JH AH JH ] 19 Glides Liquids Nasals Stops Fricatives Affricates a This set of 39 phonemes is used in the CMU Pronouncing Dictionary available on-line at http://www.1 Condensed list of ARPAbet phonetic symbols for North American English. We see that.cs.1 Phonetic Representation of Speech Table 2. phonemes have a distinctive appearance in the speech waveform. 4 shows how the sounds corresponding to the text “should we chase” are encoded into a speech waveform.

1) Fig. tongue. glides. or /NG/.1 in several ways. This branch path radiates sound at the nostrils. These differences result from the distinctively different ways that these sounds are produced.20 The Speech Signal random noise. 2. [35]. jaw. and velum. is connected to the main acoustic branch by the trapdoor action of the velum.1 Schematic model of the vocal tract system. while the vowel sounds /UH/. For creating nasal sounds like /M/.1 [35]. nasals in Table 2. called the nasal tract.2 Models for Speech Production A schematic longitudinal cross-sectional drawing of the human vocal tract mechanism is given in Figure 2. and /EY/ are highly structured and quasi-periodic. 2. Although the actual human vocal tract is not laid out along a straight line as in Figure 2. The sounds of speech are generated in the system of Figure 2. (After Flanagan et al. this type of model is a reasonable approximation for wavelengths of the sounds in speech. a side-branch tube. liquids. Voiced sounds (vowels. The shape (variation of cross-section along the axis) of the vocal tract varies with time due to motions of the lips. This diagram highlights the essential physical features of human anatomy that enter into the final stages of the speech production process. This tube serves as an acoustic transmission system for sounds generated inside the vocal tract. It shows the vocal tract as a tube of nonuniform cross-sectional area that is bounded at one end by the vocal cords and at the other by the mouth opening.1. /IY/. /N/.) .

/DH/. the fine structure of the time waveform is created by the sound sources in the vocal tract.2 Models for Speech Production 21 are produced when the vocal tract tube is excited by pulses of air pressure resulting from quasi-periodic opening and closing of the glottal orifice (opening between the vocal cords). All these excitation sources create a wide-band excitation signal to the vocal tract tube. Sounds produced in this manner include the voiced fricatives /V/. In summary. while the detailed time variations of the speech waveform are at a much higher rate. and /K/ and affricates such as /CH/ are formed by momentarily closing off air flow. A third sound production mechanism is when the vocal tract is partially closed off causing turbulent flow due to the constriction. and the liquid consonant /W/. These resonance frequencies are called the formant frequencies of the sound [32. the changes in vocal tract configuration occur relatively slowly compared to the detailed time variation of the speech signal. allowing pressure to build up behind the closure. the general character of the speech signal varies at the phoneme rate. and the resonances of the vocal tract tube shape these sound sources into the phonemes.1 are the vowels /UH/. The sounds created in the vocal tract are shaped in the frequency domain by the frequency response of the vocal tract. which acts as a random noise excitation of the vocal tract tube. Unvoiced sounds are produced by creating a constriction somewhere in the vocal tract tube and forcing air through that constriction. /IY/. That is. As discussed in Chapter 1 and illustrated by the waveform in Figure 1.1 can be described by acoustic theory. and /ZH/. plosive sounds such as /P/. /Z/. Examples are the unvoiced fricative sounds such as /SH/ and /S/. 34]. The system of Figure 2.2.1. Finally. Examples in Figure 1. thereby creating turbulent air flow. which is on the order of 10 phonemes per second. which acts as an acoustic transmission line with certain vocal tract shape-dependent resonances that tend to emphasize some frequencies of the excitation relative to others. and then abruptly releasing the pressure. and /EY/. The resonance frequencies resulting from a particular configuration of the articulators are instrumental in forming the sound corresponding to a given phoneme. at the same time allowing quasi-periodic flow due to vocal cord vibrations. and numerical techniques can be used to create a complete physical . /T/.

93]. Samples of a speech signal are assumed to be the output of the time-varying linear system.2) change at a rate on the order of 50–100 times/s.2. (2. The excitation generator on the left simulates the different modes of sound generation in the vocal tract. The short-time frequency response of the linear system simulates the frequency shaping of the vocal tract system. Thus. 64]. In detailed modeling of speech production [32. Some of the poles (ck ) of the system function lie close to the unit circle and create resonances to model the formant frequencies. The discrete-time time-varying linear system on the right in Figure 2. it is reasonable to assume that the linear system response does not vary over time intervals on the order of 10 ms or so.2 simulates the frequency shaping of the vocal tract tube. simulation of sound generation and transmission in the vocal tract [36. 2. it is common to characterize the discretetime linear system by a system function of the form: M bk z H(z) = k=0 N −k M b0 = k=1 N k=1 (1 − dk z −1 ) . it is sufficient to model the production of a sampled speech signal by a discrete-time system model such as the one depicted in Figure 2.2 Source/system model for a speech signal. and since the vocal tract changes shape relatively slowly. In general such a model is called a source/system model of speech production. for the most part. 34.22 The Speech Signal Excitation Parameters Excitation Generator excitation signal Vocal Tract Parameters Linear System speech signal Fig. but.1) 1− k=1 ak z −k (1 − ck z −1 ) where the filter coefficients ak and bk (labeled as vocal tract parameters in Figure 2. it is sometimes useful to employ zeros (dk ) of the system function to model nasal and fricative .

This model of speech as the output of a slowly time-varying digital filter with an excitation that captures the nature of the voiced/unvoiced distinction in speech production is the basis for thinking about the speech signal. it is possible to compute/measure/estimate the parameters of the model by analyzing short blocks of samples of the speech signal. That is. For voiced speech the excitation to the linear system is a quasi-periodic sequence of discrete (glottal) pulses that look very much like those shown in the righthand half of the excitation signal waveform in Figure 2. It is through such models and analysis techniques that we are able to build properties of the speech production process into digital representations of the speech signal. Often it is convenient to absorb the glottal pulse spectrum contribution into the vocal tract system model of (2. In either case.2 creates an appropriate excitation for the type of sound being produced. the linear system imposes its frequency response on the spectrum to create the speech sounds.2. and a wide variety of digital representations of the speech signal are based on it.2 Models for Speech Production 23 sounds.2. Therefore.2 switches from unvoiced to voiced leading to the speech signal output as shown in the figure. The fundamental frequency of the glottal excitation determines the perceived pitch of the voice.1). By assuming that the properties of the speech signal (and the model) are constant over short time intervals. The box labeled excitation generator in Figure 2. as we discuss further in Chapter 4. For unvoiced speech. The individual finite-duration glottal pulses have a lowpass spectrum that depends on a number of factors [105]. the speech signal is represented by the parameters of the model instead of the sampled waveform. This can be achieved by a small increase in the order of the denominator over what would be needed to represent the formant resonances. the linear system is excited by a random number generator that produces a discrete-time noise signal with flat spectrum as shown in the left-hand half of the excitation signal. The excitation in Figure 2. However. . many applications of the source/system model only include poles in the model because this simplifies the analysis required to estimate the parameters of the model from the speech signal. the periodic sequence of smooth glottal pulses has a harmonic line spectrum with components that decrease in amplitude with increasing frequency.

51] much work has been devoted to creating detailed simulations of glottal flow. Stevens [121] and Quatieri [94] provide useful discussions of these effects.24 The Speech Signal 2. such models are based on many approximations including the assumption that the source and the system do not interact. Since the early work of Flanagan and Ishizaka [34. . Early efforts to measure vocal tract area functions involved hand tracing on X-ray pictures [32]. researchers have sought to measure the physical dimensions of the human vocal tract during speech production.3 More Refined Models Source/system models as shown in Figure 2. Recent advances in MRI imaging and computer image analysis have provided significant advances in this area of speech science [17]. However. For many years. 36. and we shall rely on such models throughout this text. the assumption of linearity. and the assumption that the distributed continuous-time vocal tract transmission system can be modeled by a discrete linear time-invariant system. Fluid mechanics and acoustic wave propagation theory are fundamental physical principles that must be applied for detailed modeling of speech production. This information is essential for detailed simulations based on acoustic theory. and the nonlinearities that enter into sound generation and transmission in the vocal tract.2 with the system characterized by a time-sequence of time-invariant systems are quite sufficient for most applications in speech processing. the interaction of the glottal source and the vocal tract in speech production.

which consists of the cochlea and the set of neural connections to the auditory nerve. the middle ear beginning at the tympanic membrane. and finally.3 Hearing and Auditory Perception In Chapter 2.1 The Human Ear Figure 3. we introduced the speech production process and showed how we could model speech production using discrete-time systems. and including three small bones.1 shows a schematic view of the human ear showing the three distinct sound processing sections. the incus (also called the anvil) and the stapes (also called the stirrup). or eardrum. the malleus (also called the hammer). 3. the inner ear. which perform a transduction from acoustic waves to mechanical pressure waves. namely: the outer ear consisting of the pinna. which gathers sound and conducts it through the external canal to the middle ear. In this chapter we turn to the perception side of the speech chain to discuss properties of human sound perception that can be employed to create digital representations of the speech signal that are perceptually robust. which conducts the neural signals to the brain. 25 .

26 Hearing and Auditory Perception Fig. (After Flanagan [34]. The basilar membrane vibrates in a frequency-selective manner along its extent and thereby performs a rough (non-uniform) spectral analysis of the sound. [107].) Figure 3. The acoustic wave is transmitted from the outer ear to the inner ear where the ear drum and bone structures convert the sound wave to mechanical vibrations which ultimately are transferred to the basilar membrane inside the cochlea. Distributed along Fig.1 Schematic view of the human ear (inner and middle structures enlarged).2 [107] depicts a block diagram abstraction of the auditory system. (After Sachs et al.2 Schematic model of the auditory mechanism.) . 3. 3.

2 Perception of Loudness A key factor in the perception of speech and other sounds is loudness. For example. Even so. the dotted curve at the bottom of the figure labeled “threshold of audibility” shows the sound pressure level that is required for a sound of a given frequency to be just audible (by a person with normal hearing). The solid curves are equal-loudness-level contours measured by comparing sounds at various frequencies with a pure tone of frequency 1000 Hz and known sound pressure level. shown in Figure 3. Loudness is a perceptual quality that is related to the physical property of sound pressure level.2 Perception of Loudness 27 the basilar membrane are a set of inner hair cells that serve to convert motion along the basilar membrane to neural activity. This produces an auditory nerve representation in both time and frequency. Careful measurements of this kind show that a 100 Hz tone must have a sound pressure level of about 60 dB in order .3 [37. 103]. The processing at higher levels in the brain. These loudness curves show that the perception of loudness is frequency-dependent. Specifically.3.2 as a sequence of central processing with multiple representations followed by some type of pattern recognition. Loudness is quantified by relating the actual sound pressure level of a pure tone (in dB relative to a standard reference level) to the perceived loudness of the same tone (in a unit called phons) over the range of human hearing (20 Hz–20 kHz). These experiments have yielded much valuable knowledge about the sensitivity of the human auditory system to acoustic properties such as intensity and frequency. a wealth of knowledge about how sounds are perceived has been discovered by careful experiments that use tones and noise signals to stimulate the auditory system of human observers in very specific and controlled ways. the point at frequency 100 Hz on the curve labeled 50 (phons) is obtained by adjusting the power of the 100 Hz tone until it sounds as loud as a 1000 Hz tone having a sound pressure level of 50 dB. is not well understood and we can only postulate the mechanisms used by the human brain to perceive sound or speech. 3. This relationship is shown in Figure 3. It can be seen that low frequencies must be significantly more intense than frequencies in the mid-range in order that they be perceived at all.

4.28 Hearing and Auditory Perception Fig. the bandpass filters are not ideal .3 Critical Bands The non-uniform frequency analysis performed by the basilar membrane can be thought of as equivalent to that of a set of bandpass filters whose frequency responses become increasingly broad with increasing frequency. In reality. By convention. 3. (After Fletcher and Munson [37].) to be perceived to be equal in loudness to the 1000 Hz tone of sound pressure level 50 dB. An idealized version of such a filter bank is depicted schematically in Figure 3.3 Loudness level for human hearing. The equal-loudness-level curves show that the auditory system is most sensitive for frequencies ranging from about 100 Hz up to about 6 kHz with the greatest sensitivity at around 3 to 4 kHz. 3. both the 50 dB 1000 Hz tone and the 60 dB 100 Hz tone are said to have a loudness level of 50 phons (pronounced as /F OW N Z/). This is almost precisely the range of frequencies occupied by most of the sounds of speech.

and with a relative bandwidth of about 20% of the center frequency above 500 Hz. The concept of critical bands is very important in understanding such phenomena as loudness perception.4 Schematic representation of bandpass filters according to the critical band theory of hearing. An equation that fits empirical measurements over the auditory range is ∆fc = 25 + 75[1 + 1. which is a physical attribute of the acoustic waveform [122].4(fc /1000)2 ]0.4.4 Pitch Perception amplitude 29 Fig. Approximately 25 critical band filters span the range from 0 to 20 kHz. (3.1) where ∆fc is the critical bandwidth associated with center frequency fc [134]. pitch perception.4 Pitch Perception Most musical sounds as well as voiced speech sounds have a periodic structure when viewed over short time intervals.3. as shown in Figure 3. and such sounds are perceived by the auditory system as having a quality known as pitch. Like loudness. 3. Even so. the concept of bandpass filter analysis in the cochlea is well established. showing that the effective bandwidths are constant at about 100 Hz for center frequencies below 500 Hz. and the critical bandwidths have been defined and measured using a variety of methods.69 . and masking. 3. but their frequency responses overlap significantly since points on the basilar membrane cannot vibrate independently of each other. The relationship between pitch (measured . pitch is a subjective attribute of sound that is related to the fundamental frequency of the sound. and it therefore provides motivation for digital representations of the speech signal that are based on a frequency decomposition.

This empirical scale describes the results of experiments where subjects were asked to adjust the pitch of a measurement tone to half the pitch of a reference tone. For example.2) which is plotted in Figure 3. (3. Below 1000 Hz.5. 3. can be related to the concept of critical bands [134]. To calibrate the scale.2) shows that a frequency of f = 5000 Hz corresponds to a pitch of 2364 mels. This is shown in Figure 3. however. the relationship is nonlinear. This expression is calibrated so that a frequency of 1000 Hz corresponds to a pitch of 1000 mels.5 Relation between subjective pitch and frequency of a pure tone. what we know about pitch perception reinforces the notion that the auditory system performs a frequency analysis that can be simulated with a bank of bandpass filters whose bandwidths increase as center frequency increases. one critical bandwidth corresponds to about 100 mels on the pitch scale. on a nonlinear frequency scale called the mel-scale) and frequency of a pure tone is approximated by the equation [122]: Pitch in mels = 1127 loge (1 + f /700). where a critical band of width ∆fc = 160 Hz centered on fc = 1000 Hz maps into a band of width 106 mels and a critical band of width 100 Hz centered on 350 Hz maps into a band of width 107 mels.5. as quantified by the melscale. The psychophysical phenomenon of pitch. the relationship between pitch and frequency is nearly proportional.30 Hearing and Auditory Perception Fig. It turns out that more or less independently of the center frequency of the band. (3. a tone of frequency 1000 Hz is given a pitch of 1000 mels. Thus. For higher frequencies. .

but contains many frequencies. an intense tone (called the masker) tends to raise the threshold of audibility around its location on the frequency axis as shown by the solid line. 3. A related phenomenon. Masking occurs when one sound makes a second superimposed sound inaudible. Often the term pitch period is used for the fundamental period of the voiced speech signal even though its usage in this way is somewhat imprecise.6 Illustration of effects of masking. Nevertheless. Figure 3. 3. .6. As shown in Figure 3.3. A detailed discussion of masking can be found in [134]. and noise can mask pure tones as well. is also attributable to the mechanical vibrations of the basilar membrane. many of the results obtained with pure tones are relevant to the perception of voice pitch as well.6 illustrates masking of tones by tones. Loud tones causing strong vibrations at a point on the basilar membrane can swamp out vibrations that occur nearby. Pure tones can mask other pure tones.5 Auditory Masking 31 Voiced speech is quasi-periodic.5 Auditory Masking The phenomenon of critical band auditory analysis can be explained intuitively in terms of vibrations of the basilar membrane. called masking. The notion that a sound becomes inaudible can be quantified with respect to the threshold of audibility. All spectral components whose level is below this raised threshold are masked and therefore do amplitude Masker Shifted threshold Unmasked signal Masked signals (dashed) Threshold of hearing log frequency Fig.

.32 Hearing and Auditory Perception not need to be reproduced in a speech (or audio) processing system because they would not be heard.6 Complete Model of Auditory Processing In Chapter 2. Masking is widely employed in digital representations of speech (and audio) signals by “hiding” errors in the representation in areas where the threshold of hearing is elevated by strong frequency components in the signal [120]. In this chapter. 48. and thus we rely on psychophysical experimentation to understand the role of loudness. Similarly. it is possible to achieve lower data rate representations while maintaining a high degree of perceptual fidelity. we described elements of a generative model of speech production which could. beyond the basilar membrane processing of the inner ear. completely describe and model the ways in which speech is produced by humans. critical bands. in theory. pitch perception.6 where the falloff of the shifted threshold is less abrupt above than below the masker. This is shown in Figure 3. 115] and used in a range of speech processing systems. is rudimentary at best. we described elements of a model of speech perception. and therefore will be heard. Although some excellent auditory models have been proposed [41. It has been shown that the masking effect is greater for frequencies above the masking frequency than below. the problem that arises is that our detailed knowledge of how speech is perceived and understood. 3. all such models are incomplete representations of our knowledge about how speech is understood. In this way. 73. However. and auditory masking in speech perception in humans. any spectral component whose level is above the raised threshold is not masked.

and the voiced excitation is assumed to be a periodic impulse train with impulses spaced by the pitch period (P0 ) rounded to the nearest 33 . we presented a model for speech production in which an excitation source provides the basic temporal fine structure while a slowly varying filter provides spectral shaping (often referred to as the spectrum envelope) to create the various sounds of speech. and few details of the excitation or the linear system were given. In Figure 2. Since our goal is to extract parameters of the model by analysis of the speech signal. the source/system separation was presented at an abstract level. One such model uses a more detailed representation of the excitation in terms of separate source generators for voiced and unvoiced speech as shown in Figure 4.2.4 Short-Time Analysis of Speech In Figure 2.2 of Chapter 2. In this model the unvoiced excitation is assumed to be a random noise sequence. this is not sufficient to uniquely specify either the excitation or the system.1. Clearly. Both the excitation and the linear system were defined implicitly by the assertion that the sampled speech signal was the output of the overall system. it is common to assume structures (or representations) for both the excitation generator and the linear system.

and system function of the system remains relatively constant. We use n for the time index within that interval.1). . it is often assumed that the system is an all-pole system with 1 As mentioned in Chapter 3. the impulse response. which is assumed to be slowly-time-varying (changing every 50–100 ms or so). ˆ ˆ the most general linear time-invariant system would be characterized by a rational system function as given in (2. and m is ˆ the index of summation in the convolution sum. the gain Gn is absorbed into hn [m] for convenience.34 Short-Time Analysis of Speech Fig.1 The pulses needed to model the glottal flow waveform during voiced speech are assumed to be combined (by convolution) with the impulse response of the linear system. In this model.1 Voiced/unvoiced/system model for a speech signal. For example over time intervals of tens of milliseconds. As discussed in Chapter 2. frequency response.1) where the subscript n denotes the time index pointing to the block of ˆ samples of the entire speech signal s[n] wherein the impulse response hn [m] applies. the system can be described by the convolution expression ∞ sn [n] = ˆ m=0 hn [m]en [n − m]. the period of voiced speech is related to the fundamental frequency (perceived as pitch) of the voice. 4. By this we mean that over the timescale of phonemes. ˆ ˆ (4. to simplify analysis. However. sample.

we use n and m for discrete indices for sequences. we use n. the linear system in the model is commonly referred to as simply the “vocal tract” system and the corresponding impulse response is called the “vocal tract impulse response. Because of the slowly varying nature of the speech signal. This leads to the basic principle of short-time analysis.2 Although the linear system is assumed to model the composite spectrum effects of radiation. but whenever we want to indicate a specific analysis time. and techniques for extracting the linear system model (which imparts the spectrum envelope or spectrum shaping). but ˆ this complication of notation is not generally necessary since it is usually clear that the system function or difference equation only applies over a short time interval. which 2 In general. such as the pitch period and voiced/unvoiced classification.3) where as discussed above.2) The coefficients G and ak in (4.2) change with time. ak z −k (4. and glottal excitation pulse shape (for voiced speech only) over a short time interval.35 system function of the form: H(z) = 1− k=1 G p .1. common practice is to partition the analysis of the speech signal into techniques for extracting the parameters of the excitation model. vocal tract tube.” For all-pole linear systems. and should therefore be indexed with the subscript n as in (4.2). With such a model as the basis. we have suppressed the indication of the time at which the difference equation applies. ˆ .1) and Figure 4. it is common to process speech in blocks (also called “frames”) over which the properties of the speech waveform can be assumed to remain relatively constant. the input and output are related by a difference equation of the form: p s[n] = k=1 ak s[n − k] + Ge[n]. as represented by (4. (4.

they will be concentrated in time and frequency.. ∞ m=−∞ w[m]. n For example. a finite-duration4 window might be a Hamming window defined by wH [m] = 0. Also shown is a sequence of data windows of duration 40 ms and shifted by 15 ms (320 samples at 16 kHz sampling rate) between windows. 5 We have assumed the time origin at the center of a symmetric interval of 2M + 1 samples.36 Short-Time Analysis of Speech is represented in a general form by the equation: ∞ Xn = ˆ m=−∞ T {x[m]w[ˆ − m]} . ˆ 4 An example of an infinite-duration window is w[n] = nan for n ≥ 0. as used throughout this chapter. for all m in the region of support of the window. The infinite limits in (4. (4. i. .3 The operator T { } defines the nature ˆ of the short-time analysis function.5 It can be shown that a (2M + 1) sample Hamming window has a frequency main lobe (full) bandwidth of 4π/M .e. This illustrates how short-time analysis is implemented. and the frequency width will be inversely proportional to the time width [89]. We will see several ˆ examples of such operators that are designed to extract or highlight certain features of the speech signal. Figure 4. i.. 3 Sometimes (4.e. i.3 shows a 125 ms segment of a speech waveform that includes both unvoiced (0–50 ms) and voiced speech (50–125 ms).4) where Xn represents the short-time analysis parameter (or vector of ˆ parameters) at analysis time n. whose purpose is to select a segment of the sequence x[m] in the neighborhood of sample m = n. Such a window can lead to recursive implementations of short-time analysis functions [9]. A causal window would be shifted to the right by M samples.4) is normalized by dividing by the effective window length.4) imply summation over all nonzero values of the windowed segment xn [m] = ˆ x[m]w[ˆ − m]. and w[ˆ − m] represents a timen shifted window sequence. so that Xn is a weighted average.46 cos(πm/M ) 0 −M ≤ m ≤ M otherwise. Other windows will have similar properties.54 + 0.2 shows a discrete-time Hamming window and its discrete-time Fourier transform. n (4..e.5) Figure 4.

1 Short-Time Energy and Zero-Crossing Rate Two basic short-time analysis functions useful for speech signals are the short-time energy and the short-time zero-crossing rate. 4. The short-time energy is defined as ∞ ∞ En = ˆ m=−∞ (x[m]w[ˆ − m]) = n m=−∞ 2 x2 [m]w2 [ˆ − m].4. and they are useful for estimating properties of the excitation function in the model.2 Hamming window (a) and its discrete-time Fourier transform (b). These functions are simple to compute.3 Section of speech waveform with short-time analysis windows. 4.6) . n (4. 4. 0 20 40 60 time in ms 80 100 120 Fig.1 Short-Time Energy and Zero-Crossing Rate (a) 37 (b) Fig.

the computation of Zn could be implemented in ˆ other ways. En = x2 [n] ∗ he [n] n=ˆ .2(b). it follows that Zn in (4.4). Figure 4.5|sgn{x[m]} − sgn{x[m − 1]}| is equal to 1 if x[m] and x[m − 1] have different algebraic signs and 0 if they have the same sign. both the short-time energy and the short-time zero-crossing rate are output of a lowpass filter whose frequency response is as shown in Figure 4. n (4. the frequency response is very small for discrete-time frequencies above 2π/200 rad/s (equivalent to 16000/200 = 80 Hz analog frequency).5).4. In both cases. Similarly. For the 401-point Hamming window used in Figure 4.6).7) where sgn{x} = 1 −1 x≥0 x < 0.6 Thus. so that he [n] = w2 [n] is the Hamming window defined by (4. the window is a Hamming window (two examples shown) of duration 25 ms (equivalent to 401 samples at a 16 kHz sampling rate).38 Short-Time Analysis of Speech In this case the operator T { } is simply squaring the windowed samples. where the impulse response of the linear filter ˆ n is he [n] = w2 [n]. Representing this operator in terms of linear filtering leads to ∞ Zn = ˆ m=−∞ 0. (4. While this is a convenient representation that fits the general n framework of (4. it is often possible to express short-time analysis operators as a convolution or linear filtering operation. This means that the short-time 6 In the case of the short-time energy.4 shows an example of the short-time energy and zerocrossing rate for a segment of speech with a transition from unvoiced to voiced speech.8) Since 0.7) is a weighted sum of all the instances of alternating sign ˆ (zero-crossing) that fall within the support region of the shifted window w[ˆ − m]. the short-time zero crossing rate is defined as the weighted average of the number of times the speech signal changes sign within the time window.5 |sgn{x[m]} − sgn{x[m − 1]}| w[ˆ − m]. As shown in (4. the window applied to the signal samples was the square-root of the Hamming window. In this case. .

4. they can be sampled at a much lower rate than that of the original speech signal. From ˆ our model. The short-time energy and short-time zero-crossing rate are important because they abstract valuable information about the speech signal. we expect unvoiced regions to have lower short-time energy than voiced regions. the zero-crossing rate is relatively high compared to the zero-crossing rate in the voiced interval. Similarly. The short-time energy is an indication of the amplitude of the signal in the interval around time n. this reduction of the sampling rate is accomplished by moving the window position n in ˆ jumps of more than one sample as shown in Figure 4. This is due to the time delay of M samples (equivalent to 12.4 Section of speech waveform with short-time energy and zero-crossing rate superimposed. Note that during the unvoiced interval.5 ms) added to make the analysis window filter causal. the short-time energy and short-time zero-crossing rate can be the basis for an algorithm for making a decision as to whether the speech signal is voiced or unvoiced at . the short-time zero-crossing rate is a crude frequency analyzer. while unvoiced sounds have much more HF energy. Conversely.4. Thus.1 Short-Time Energy and Zero-Crossing Rate 39 0 20 40 60 time in ms 80 100 120 Fig. energy and zero-crossing rate functions are slowly varying compared to the time variations of the speech signal. For finite-length windows like the Hamming window.3. Voiced signals have a high frequency (HF) falloff due to the lowpass nature of the glottal pulses. the energy is relatively low in the unvoiced region compared to the energy in the voiced region. and therefore. Note also that there is a small shift of the two curves relative to events in the time waveform. and they are simple to compute.

1 where . If the window has finite duration.9) Using the familiar even-symmetric property of the autocorrelation.9) can be expressed in terms of linear time-invariant ˆ ˆ (LTI) filtering as ∞ φn [ ] = ˆ m=−∞ x[m]x[m − ]w [ˆ − m]. The STACF is defined as the deterministic autocorrelation function of the n sequence xn [m] = x[m]w[ˆ − m] that is selected by the window shifted ˆ to time n. and it is also the basis for many spectrum analysis methods. These distributions can be used to derive thresholds used in voiced/unvoiced decision [100].6).9) can be evaluated directly or using FFT techniques (see Section 4. ˆ ∞ φn [ ] = ˆ m=−∞ ∞ xn [m]xn [m + ] ˆ ˆ x[m]w[ˆ − m]x[m + ]w[ˆ − m − ]. This makes it a useful tool for short-time speech analysis. ˜ n (4. the short-time autocorrelation of (4. 9]. (4. φn [− ] = φn [ ]. For infinite-duration decaying exponential windows. n n m=−∞ = (4.. ˜ To see how the short-time autocorrelation can be used in speech analysis. A complete algorithm would involve measureˆ ments of the statistical distributions of the energy and zero-crossing rate for both voiced and unvoiced speech segments (and also background noise distributions).e. (4.10) where w [m] = w[m]w[m + ]. 4.2 Short-Time Autocorrelation Function (STACF) The autocorrelation function is often used as a means of detecting periodicity in signals. i.40 Short-Time Analysis of Speech any particular time n. assume that a segment of the sampled speech signal is a segment of the output of the discrete-time model shown in Figure 4.10) can be computed recursively at time n by using a ˆ different filter w [m] for each lag value [8. Note that the STACF is a two˜ dimensional function of the discrete-time index n (the window position) ˆ and the discrete-lag index .

the excitation can be assumed to be random white noise.2 Short-Time Autocorrelation Function (STACF) 41 the system is characterized at a particular analysis time by an impulse response h[n]. the autocorrelation of a periodic impulse train excitation with period P0 is a periodic impulse train sequence with the same period. That is. and radiation at the lips. In the case of the speech signal. therefore. the deterministic autocorrelation function of a finite-length segment of the speech waveform will have properties similar to those of (4. However. and the input is either a periodic impulse train or random white noise. the autocorrelation function of unvoiced speech computed using averaging would be simply φ(s) [ ] = φ(h) [ ].) Different segments of the speech signal will have the same form of model with different excitation and system impulse response. whose stochastic autocorrelation function would be an impulse sequence at = 0. and (4.12) In the case of unvoiced speech. where e[n] is the excitation to the linear system with impulse response h[n]. the autocorrelation of voiced speech is the periodic autocorrelation function ∞ φ (s) [ ]= m=−∞ φ(h) [ − mP0 ].4. For voiced speech. Therefore. the autocorrelation function of s[n] = e[n] ∗ h[n] is the convolution of the autocorrelation functions of e[n] and h[n]. vocal tract shape.12) and (4. (Note that we have suppressed the indication of analysis time. (4. and easily proved. In this case. assume that s[n] = e[n] ∗ h[n]..e.13) Equation (4. h[n] represents the combined (by convolution) effects of the glottal pulse shape (for voiced speech). A well known.12) assumes periodic computation of an infinite periodic signal.13) except that the correlation values will taper off with lag due to the tapering of the window and the fact that less and less data is involved in the computation of the short-time autocorrelation for longer . property of the autocorrelation function is that φ(s) [ ] = φ(e) [ ] ∗ φ(h) [ ].11) i. (4. (4.13) assumes probability average or averaging over an infinite time interval (for stationary random signals).

2 0.5 0 10 20 30 40 (c) Unvoiced Segment (d) Unvoiced Autocorrelation 0.5 Voiced and unvoiced segments of speech and their corresponding STACF.3 Short-Time Fourier Transform (STFT) The short-time analysis functions discussed so far are examples of the general short-time analysis principle that is the basis for most .14) 4.2 0. lag values.2 0 10 20 time in ms 30 40 1 0. This tapering off in level is depicted in Figure 4.5 0 20 time in ms 40 Fig.42 Short-Time Analysis of Speech (a) Voiced Segment (b) Voiced Autocorrelation 0. This suggests that the STACF could be the basis for an algorithm for estimating/detecting the pitch period of speech.5 0 −0.1 0 −0.5 0 −0. observe that the STACF implicitly contains the short-time energy since ∞ En = ˆ m=−∞ (x[m]w[ˆ − m])2 = φn [0].2 0 10 20 30 40 1 0.1 −0. Finally. 4.5 for both a voiced and an unvoiced speech segment. Usually such algorithms involve the autocorrelation function and other short-time measurements such as zero-crossings and energy to aid in making the voiced/unvoiced decision.1 −0. Note the peak in the autocorrelation function for the voiced segment at the pitch period and twice the pitch period. n ˆ (4.1 0 −0. and note the absence of such peaks in the autocorrelation function for the unvoiced segment.

By definition.e. (4.16) or. we will find that the STFT defined as ∞ Xn (e ) = ˆ m=−∞ jω ˆ ˆ x[m]w[ˆ − m]e−j ωm ..15) can be expressed as the discrete convolution ˆ ˆ Xn (ej ω ) = (x[n]e−j ωn ) ∗ w[n] ˆ n=ˆ n . we use n to specify the analysis time. 89. alternatively. the DTFT of the (usually finite-duration) signal selected and amplitude-weighted by the sliding window w[ˆ − m] [24.15) is the basis for a wide range of speech analysis. Xn (ej ω ) ˆ ˆ is slowly varying as n varies. .15) is a sequence of DTFTs. In subsequent chapters. the STFT is the discreteˆ n time Fourier transform (DTFT) of the signal xn [m] = x[m]w[ˆ − m]. ˆ As in the case of the other short-time analysis functions discussed in this chapter.3 Short-Time Fourier Transform (STFT) 43 algorithms for speech processing. Thus. coding and synthesis systems.2(b). (4.7 ˆ Since (4. for fixed analysis time n. and ω is used to distinguish the STFT ˆ ˆ analysis frequency from the frequency variable of the non-time-dependent Fourier transform frequency variable ω. n the discrete-time index denotˆ ing the window position. ˆ ˆ ˆ Xn (ej ω ) = x[n] ∗ (w[n]ej ωn ) e−j ωn ˆ n=ˆ n .4. has a lowpass frequency response with the cutoff frequency varying inversely with the window length. 129]. the STFT can be expressed in terms of a linear filtering operation.) This means that for a fixed value of ω . and ω representing the analysis frequency. Equation (4. the two-dimensional function ˆ ˆ Xn (ej ω ) at discrete-time n is a periodic function of continuous radian ˆ frequency ω with period 2π [89].16) can be interpreted as ˆ ˆ follows: the amplitude modulation x[n]e−j ωn shifts the spectrum of 7 As before. For example. ˆ (See Figure 4. (4. when viewed as a linear filter impulse response. n the STFT is a function of two variables. n (4.17) Recall that a typical window like a Hamming window. We now turn our attention to what is perhaps the most important basic concept in digital speech processing. ˆ i.

. N − 1. which selects the band of frequencies centered on ω . the band of frequencies of x[n] that were originally centered on analysis frequency ω . The STFT becomes a practical tool for both analysis ˆ and applications when it is implemented with a finite-duration window moved in steps of R > 1 samples in time and computed at a discrete set of frequencies as in rR XrR [k] = m=rR−L+1 x[m]w[rR − m]e−j(2πk/N )m k = 0. . This is. . Note ˆ that we have assumed that w[m] is causal and nonzero only in the range 0 ≤ m ≤ L − 1 so that the windowed segment x[m]w[rR − m] is nonzero over rR − L + 1 ≤ m ≤ rR. resulting again in the same lowpass signal [89]. and the window (lowpass) filter selects the resulting ˆ band of frequencies around zero frequency.18) in the equivalent form: ˜ XrR [k] = XrR [k]e−j(2πk/N )rR k = 0. (3) For each frequency ω . (4. An identical conclusion follows from (4. 4.15).. it is helpful to write (4. In summary. the STFT is a function of a continuous analysis frequency ω .19) .17): x[n] is the ˆ ˆ input to a bandpass filter with impulse response w[n]ej ωn . of course. it is the time sequence output of a lowpass filter that ˆ ˆ follows frequency down-shifting by ω . Then that band of frequencies is ˆ ˆ shifted down by the amplitude modulation with e−j ωn . a periodic function of ω at each window position n. the STFT has three interpretations: (1) It is a sequence of discrete-time Fourier transforms of windowed signal segments.e. N − 1. (4. it is ˆ ˆ the time sequence output resulting from frequency down-shifting the output of a bandpass filter. (2) For each frequency ˆ ˆ ω with n varying. . 1.18) where N is the number of uniformly spaced frequencies across the interval 0 ≤ ω < 2π. . To aid in interpretation. 1.44 Short-Time Analysis of Speech x[n] down by ω . i. .4 Sampling the STFT in Time and Frequency As defined in (4. . and L is the window length (in samples). .

N − 1. and the segment of the speech signal is the time-reversed sequence of L samples that precedes the analysis time. 1. used to compute the STFT [1]. due to the definition of the window. for m = 0. (4.) (3) The multiplication by e−j(2πk/N )rR can be done if necessary. has the interpretation of an N -point DFT of the sequence x[rR − m]w[m]. Thus. L − 1. (4. that w[m] = 0 only in the range ˜ 0 ≤ m ≤ L − 1. 1.4 Sampling the STFT in Time and Frequency 45 where L−1 ˜ XrR [k] = m=0 x[rR − m]w[m]ej(2πk/N )m k = 0. is nonzero in the interval 0 ≤ m ≤ L − 1. . . .20). and the number of uniform frequencies. the analysis time rR is shifted to the time origin of the DFT computation. . and (3). the alternative form.4. (2) Compute the complex conjugate of the N -point DFT of the sequence xrR [m]. (This can be done efficiently with an N -point FFT algorithm.e. giving the 8 The DFT is normally defined with a negative exponent. (4) Move the time origin by R samples (i. for specificity. . but often can be omitted (as in computing a sound spectrogram or spectrographic display). . since x[rR − m]w[m] is real. The complex exponential factor e−j(2πk/N )rR in (4. . The remaining issue for complete specification of the sampled STFT is specification of the temporal sampling period. It can easily be shown that both R and N are determined entirely by the time width and frequency bandwidth of the lowpass window.19) results from the shift of the time origin. it follows that ˜ XrR [k] can be computed by the following process: (1) Form the sequence xrR [m] = x[rR − m]w[m].20) Since we have assumed. R.20) is the complex conjugate of the DFT of the windowed sequence [89]. w[m].. N .8 In (4. (2). XrR [k]. . which. . r → r + 1) and repeat steps (1). etc. For a discrete-time Fourier transform interpretation.

6. where L is the window length in samples. Specifically.6 are plots of ˜ S(tr .21) where the plot axes are labeled in terms of analog time and frequency through the relations tr = rRT and fk = k/(N T ). (4.5 The Speech Spectrogram Since the 1940s. In order to make smooth looking plots like those in Figure 4. and electrically sensitive paper [66]. R is usually quite small compared to both the window length L and the number of samples in the frequency dimension. which may be much larger than the window length L. Figure 4. Constraint (1) above is related to sampling the STFT in time at a rate of twice the window bandwidth in frequency in order to eliminate aliasing in frequency of the STFT. Such a function of two variables can be plotted on a two dimensional surface (such as this text) as either a gray-scale or a color-mapped image. C = 1 for a rectangular window. the images in Figure 4. Today spectrograms like those in Figure 4. and constraint (2) above is related to sampling in frequency at a rate of twice the equivalent time width of the window to ensure that there is no aliasing in time of the STFT. variable analog bandpass filter. Sound spectrograms like those in Figure 4. 4. spectrograms were made by an ingenious device comprised of an audio tape loop.6 are simply a display of the magnitude of the STFT. the sound spectrogram has been a basic tool for gaining understanding of how the sounds of speech are produced and how phonetic information is encoded in the speech signal. where T is the sampling period of the discrete-time signal x[n] = xa (nT ).6 shows the time .46 Short-Time Analysis of Speech following constraints on R and N : (1) R ≤ L/(2C) where C is a constant that is dependent on the window frequency bandwidth. N . (2) N ≥ L.6 are made by DSP techniques [87] and displayed as either pseudo-color or gray-scale images on computer screens. fk ) = 20 log10 |XrR [k]| = 20 log10 |XrR [k]|. C = 2 for a Hamming window. Up until the 1970s.

20) and the corresponding spectrogram images leads to valuable insight into the nature of the speech signal.6 was computed with a window length of 101 samples. The bars on the right calibrate the color map (in dB).4. The length of the window has a major effect on the spectrogram image. A careful interpretation of (4. 4. This window length is on the order of the length of a pitch period of the waveform during voiced intervals.1.5 The Speech Spectrogram 1 0 −1 5 frequency in kHz 4 3 2 SH UH D W IY CH EY S 47 0 −50 −100 1 0 0 5 100 200 300 400 500 600 0 −20 −40 −60 −80 −100 −120 frequency in kHz 4 3 2 1 0 0 100 200 300 time in ms 400 500 600 Fig. in voiced intervals. waveform at the top and two spectrograms computed with different length analysis windows. The upper spectrogram in Figure 4. then mostly small amplitude . First note that the window sequence w[m] is nonzero only over the interval 0 ≤ m ≤ L − 1. corresponding to 10 ms time duration. the spectrogram displays vertically oriented striations corresponding to the fact that the sliding window sometimes includes mostly large amplitude samples. As a result.6 Spectrogram for speech signal of Figure 1.

4.5 4 4. which is characterized by good frequency resolution and poor time resolution.7. The upper plot in Figure 4.6.5 3 3. This is consistent with the linear filtering interpretation of the STFT. the spectrogram is called a wide-band spectrogram. Conversely. each individual pitch period is resolved in the time dimension. if the analysis window is short. for example.7 Short-time spectrum at time 430 ms (dark vertical line in Figure 4.5 1 1.5 2 2. the spectrogram is a narrow-band spectrogram.5 3 frequency in kHz 3. As a result of the short analysis window.6) with Hamming window of length M = 101 in upper plot and M = 401 in lower plot. fk ) as a function of fk at time tr = 430 ms.48 Short-Time Analysis of Speech samples. For this reason. but the resolution in the frequency dimension is poor. Note the three broad peaks in the spectrum slice at time tr = 430 ms. and observe that similar slices would log magnitude (dB) 0 −20 −40 −60 −80 0 0. etc.5 1 1.5 5 log magnitude (dB) 0 −20 −40 −60 −80 0 0. when the window length is long.5 4 4. This vertical slice through the spectrogram is at the position of the black vertical line in the upper spectrogram of Figure 4.5 2 2.5 5 Fig. . since a short analysis filter has a wide passband. shows S(tr .

Multiplying by the analysis window in the time-domain results in convolution of the Fourier transform of the window with the impulses in the spectrum of the periodic signal [89]. where the local maxima of the curve are spaced at multiples of the fundamental frequency f0 = 1/T0 .6 was computed with a window length of 401 samples. .6 Relation of STFT to STACF 49 be obtained at other times around tr = 430 ms. Therefore. so over the analysis window interval it acts very much like a periodic signal.7. As a result. and at integer multiples (harmonics) of the fundamental frequency [89]. but the resolution in the frequency dimension is much better.9) is related to the STFT given by ˆ 9 Each “ripple” in the lower plot is essentially a frequency-shifted copy of the Fourier transform of the Hamming window used in the analysis. the spectrogram no longer displays vertically oriented striations since several periods are included in the window no matter where it is placed on the waveform in the vicinity of the analysis time tr . for example.7. The lower spectrogram in Figure 4. fk ) as a function of fk at time tr = 430 ms. f0 . it follows that φn [ ] given by (4.4. As a result.9 4. corresponding to 40 ms time duration. where T0 is the fundamental period (pitch period) of the signal. Periodic signals have Fourier spectra that are composed of impulses at the fundamental frequency. the striations tend to be horizontally oriented in the narrow-band case since the fundamental frequency and its harmonics are all resolved. This is evident in the lower plot of Figure 4. The lower plot in Figure 4.15) is a discrete-time Fourier transform for fixed window position. These large peaks are representative of the underlying resonances of the vocal tract at the corresponding time in the production of the speech signal. the signal is voiced at the position of the window. Since the STFT defined in (4.6 Relation of STFT to STACF A basic property of Fourier transforms is that the inverse Fourier transform of the magnitude-squared of the Fourier transform of a signal is the autocorrelation function for that signal [89]. In this case. This window length is on the order of several pitch periods of the waveform during voiced intervals. shows S(tr . the spectrogram is not as sensitive to rapid time variations.

. . Figure 4. ω ˆ (4.8 illustrates the equivalence between the STACF and the STFT. 1.5(d). Figures 4. 1.5 0 −0. .14) shows that the short-time energy can also be obtained from either (4. . but φrR [ ] = φrR [ ] for = 0.22) The STACF can also be computed from the sampled STFT. . Note that (4. L − 1 if N ≥ 2L. .15) as φn [ ] = ˆ 1 2π π −π ˆ ˆ |Xn (ej ω )|2 ej ω dˆ .23) ˜ and φrR [ ] = φrR [ ] for = 0. an inverse DFT can be used to compute 1 ˜ φrR [ ] = N N −1 ˜ |XrR (ej(2πk/N ) )|2 ej(2πk/N ) k=0 (4. . In particular. If L < N < 2L ˜ time aliasing will occur.5(b) and 4. .5 50 0 −50 −100 0 10 20 time in msec 30 40 0 2000 4000 6000 frequency in kHz 8000 Fig.23) by setting = 0.8(c) show the voiced and unvoiced autocorrelation functions that were shown in Figures 4. On the right are the corresponding STFTs.50 Short-Time Analysis of Speech (4.5 0 −0.8(a) and 4. 4.5 50 0 −50 −100 0 10 20 30 40 0 2000 4000 6000 8000 (c) Unvoiced Autocorrelation (d) Unvoiced STFT 1 0. .8 STACF and corresponding STFT.22) or (4. The peaks around 9 and 18 ms in (a) Voiced Autocorrelation (b) Voiced STFT 1 0. N − L.

19) and (4. Alternatively. and the outputs of the individual filters are called the channel signals.7 Short-Time Fourier Synthesis From the linear filtering point of view of the STFT. 4. note in the STFT on the right in Figure 4. Similarly. This structure is often called a filter bank. (4.9 shows a block diagram representation of the STFT as a combination of modulation. . 4. followed by a downsampler by R. The 2π ( 0) n N e −j X n [k ] X [k ] rR w[n] R Short Time Modifications YrR [k ] R f [n] e j 2π ( 0) n N x 2π − j (1) n e N x e R f [n] j 2π (1) n N x w[n] R x x[n] + e R f [n] j 2π ( N −1) n N y[n] e −j 2π ( N −1) n N x w[n] R x Fig. (4. The left half of Figure 4. we can estimate the fundamental frequency to be 2000/18 = 111 Hz as before.18) shows that XrR [k] is a downsampled output of the lowpass window filter with frequency-down-shifted input x[n]e−j(2πk/N )n .8(b) that there are approximately 18 local regularly spaced peaks in the range 0–2000 Hz. Thus.9 Filterbank interpretation of short-time Fourier analysis and synthesis.7 Short-Time Fourier Synthesis 51 the autocorrelation function of the voiced segment imply a fundamental frequency at this time of approximately 1000/9 = 111 Hz.20) represent the downsampled (by the factor R) output of the process of bandpass filtering with impulse response w[n]ej(2πk/N )n followed by frequency-down-shifting by (2πk/N ).4. followed by lowpass filtering.

24) and the right half of Figure 4. First. The remainder of Figure 4. defined by the part within the parenthesis in (4..9 implement the synthesis of a new time sequence y[n] from the STFT.24) The steps represented by (4. and N and careful design of the analysis window w[m] together with the synthesis window f [n]. However. An example might be quantization of the channel signals for data compression. interpolates the STFT YrR [k] to the time sampling rate of the speech signal for each channel k.52 Short-Time Analysis of Speech spectrogram is comprised of the set of outputs of the filter bank with each channel signal corresponding to a horizontal line in the spectrogram. which up-shifts the lowpass interpolated channel signals back to their original frequency bands centered on frequencies (2πk/N ). it is most important to note that with careful choice of the parameters L. . and the analysis and synthesis can be accomplished using only real filtering and with modulation operations being implicitly achieved by the downsampling [24.e. it is possible to reconstruct the signal x[n] with negligible 10 The sequence f [n] is the impulse response of a LTI filter with gain R/N and normalized cutoff frequency π/R.24).9 involve first upsampling by R followed by linear filtering with f [n] (called the synthesis window).9 shows how this can be done. The remaining parts of the structure on the right in Figure 4. There are many variations on the short-time Fourier analysis/synthesis paradigm. If we want to use the STFT for other types of speech processing. The FFT algorithm can be used to implement both analysis and synthesis. special efficiencies result when L > R = N . i. R. the diagram shows the possibility that the STFT might be modified by some sort of processing. It is often a scaled version of w[n].10 Then. from the channel signals. 129]. we may need to consider if and how it is possible to reconstruct the speech signal from the STFT. This operation. The modified STFT is denoted as YrR [k]. The sum of the up-shifted signals is the synthesized output. This part of the diagram represents the synthesis equation N −1 ∞ y[n] = k=0 r=−∞ YrR [k]f [n − rR] ej(2πk/N )n . synthesis is achieved by modulation with ej(2πk/N )n . (4.

4.25) Short-time Fourier analysis and synthesis is generally formulated with equally spaced channels with equal bandwidths. which is based on the STACF and thus equivalently on the STFT.8 Short-Time Analysis is Fundamental to our Thinking 53 error (y[n] = x[n]) from the unmodified STFT YrR [k] = XrR [k]. models for auditory processing are based on a filter bank as the first stage of processing. The fact that almost perfect reconstruction can be achieved from the filter bank channel signals gives the short-time Fourier representation major credibility in the digital speech processing tool kit. a topic beyond the scope of this text. See. Such structures are essentially the same as wavelet decompositions. This importance is strengthened by the fact that.4. is fundamental to our thinking about the speech signal and it leads us to a wide variety of techniques for achieving our goal of moving from the sampled time waveform back along the speech chain toward the implicit message. 125. we have seen that models for auditory processing involve nonuniform filter banks. and particularly the short-time Fourier representation of speech. as discussed briefly in Chapter 3. the STFT representation provides a natural framework within which this knowledge can be represented and exploited to obtain efficient representations of speech and more general audio signals. One condition that guarantees exact reconstruction is [24. (4. but one for which a large body of literature exists. but first we will consider (in Chapter 5) the technique of cepstrum analysis. Such filter banks can be implemented as tree structures where frequency bands are successively divided into low.8 Short-Time Analysis is Fundamental to our Thinking It can be argued that the short-time analysis principle. . for example [18. 129]. which is based directly on the STFT. We will have more to say about this in Chapter 7. Much of our knowledge of perceptual effects is framed in terms of frequency analysis. ∞ w[rR − n + qN ]f [n − rR] = r=−∞ 1 0 q=0 q = 0. 129].and HF bands. However. and thus. and the technique of linear predictive analysis (in Chapter 6).

was motivated by the fact that the logarithm of the Fourier spectrum of a signal containing an echo has an additive periodic component depending only on the echo size and delay.5 Homomorphic Speech Analysis The STFT provides a useful framework for thinking about almost all the important techniques for analysis of speech that have been developed so far. loosely framed in terms of spectrum analysis of analog signals. of speech. and that further Fourier analysis of the log spectrum can aid in detecting the presence 54 . the short-time cepstrum. 5. Their original definition. and Tukey to be the inverse Fourier transform of the log magnitude spectrum of a signal [16]. A more detailed discussion of the uses of the cepstrum in speech processing can be found in [110]. An important concept that flows directly from the STFT is the cepstrum. Healy.1 Definition of the Cepstrum and Complex Cepstrum The cepstrum was defined by Bogert. we explore the use of the short-time cepstrum as a representation of speech and as a basis for estimating the parameters of the speech generation model. more specifically. In this chapter.

and they extended the concept by defining the complex cepstrum as x[n] = ˆ 1 2π π log{X(ejω )}ejωn dω. (5. In the theory ˆ of homomorphic systems D∗ { } is called the characteristic system for convolution. Since we restrict our attention to real sequences. then x[n] = D∗ {x1 [n] ∗ x2 [n]} = x1 [n] + x2 [n].1a) where log |X(ejω )| is the logarithm of the magnitude of the DTFT of the signal. Schafer. c[n] = (ˆ[n] + x[−n])/2 [89]. The connection between the cepstrum concept and homomorphic filtering of convolved signals is that the complex cepstrum has the property that if x[n] = x1 [n] ∗ x2 [n].1..1b) is depicted as the block diagram in Figure 5. i. The same diagram represents the cepstrum if the complex logarithm is replaced by the logarithm of the magnitude of the DTFT. They gave a definition of the cepstrum of a discrete-time signal as c[n] = 1 2π π −π log |X(ejω )|ejωn dω. 90. 5. and Stockham showed that the cepstrum is related to the more general concept of homomorphic filtering of signals that are combined by convolution [85.1 Computing the complex cepstrum using the DTFT. x[n]. it follows from the symmetry properties of Fourier transforms that the cepstrum is the even part of the complex cepstrum. x ˆ As shown in Figure 5. the operation of computing the complex cepstrum from the input can be denoted as x[n] = D∗ {x[n]}.1b) where log{X(ejω )} is the complex logarithm of X(ejω ) defined as ˆ X(ejω ) = log{X(ejω )} = log |X(ejω )| + j arg[X(ejω )].2) Fig.1. Oppenheim.1 Definition of the Cepstrum and Complex Cepstrum 55 of that echo.e. 109]. . −π (5.5. (5.1c) The transformation implied by (5. ˆ ˆ ˆ (5.

56 Homomorphic Speech Analysis Fig. more specifically.1.3) as input. Homomorphic filtering of convolved signals is achieved by forming a modified complex cepstrum y [n] = g[n]ˆ[n]. ˆ x (5. since our model for speech production involves convolution of the excitation with the vocal tract impulse response. is what makes the cepstrum and the complex cepstrum useful for speech analysis.) which selects a portion of the complex cepstrum for inverse processing. A modified output signal y[n] can then be obtained as the output of Figure 5.3) where g[n] is a window (a “lifter” in the terminology of Bogert et al. 5. and our goal is often to separate the excitation signal from the vocal tract signal.e. i. Examples of ˆ lifters used in homomorphic filtering of speech are given in Sections 5.2 The inverse of the characteristic system for convolution (inverse complex cepstrum).2. the complex cepstrum operator transforms convolution into addition. This property. which is true for both the cepstrum and the complex cepstrum.2 with y [n] given by (5. . which shows the reverse cascade of the inverses of the operators in Figure 5. the computation of the phase angle arg[X(ejω )]. which must be done so as to preserve an additive combination of phases for two signals combined by convolution [90.2. the inverse of the characteristic system exists as in Figure 5. the output of the inverse ˆ x characteristic system will have the form y[n] = y1 [n] ∗ y2 [n]. In the case of the complex cepstrum.3) defines ˆ ˆ a linear operator in the conventional sense. 109]. Observe that (5. The key issue in the definition and computation of the complex cepstrum is the computation of the complex logarithm.4 and 5. if x[n] = x1 [n] + x2 [n] ˆ ˆ x then y [n] = g[n]ˆ1 [n] + g[n]ˆ2 [n]. where x y1 [n] = g[n]ˆ1 [n] is the complex cepstrum of y1 [n].6. Therefore. That is. etc..

but not 1 In cases where we wish to explicitly indicate analysis-time dependence of the short-time cepstrum. This is useful in the basic definitions. To emphasize this interchanging of domains of reference. The short-time cepstrum is a sequence of cepstra of windowed finite-duration segments of the speech waveform. clear. but as in ˆ other instances.2 The Short-Time Cepstrum 57 The independent variable of the cepstrum and complex cepstrum is nominally time. 5. we often suppress the subscript n. ˆ 5.1b). [16] coined the word cepstrum by transposing some of the letters in the word spectrum. and the short-time ˆ ˆ ˆ complex cepstrum is likewise defined by replacing X(ej ω ) by Xn (ej ω ) ˆ 1 The similarity to the STACF defined by (4. a “cepstrogram” would be an image obtained by plotting the magnitude of the short-time cepstrum as a function of quefrency m and analysis time n. ˆ .4) ˆ where Xn (ej ω ) is the STFT defined in (4. Only the terms cepstrum.3 Computation of the Cepstrum In (5. and liftering are widely used today.1a) and (5. The crucial observation leading to the cepstrum concept is that the log spectrum can be treated as a waveform to be subjected to further Fourier analysis. we will use n for the analysis time and m for quefrency as in (5. Thus the short-time cepstrum is defined as cn [m] = ˆ 1 2π π −π ˆ ˆ log |Xn (ej ω )|ej ωm dˆ .9) should be in (5. ω ˆ (5.1b).3). By analogy. the cepstrum and complex cepstrum are defined in terms of the DTFT.2 The Short-Time Cepstrum The application of these definitions to speech requires that the DTFT be replaced by the STFT.15).4). quefrency. They created many other special terms in this way including quefrency as the name for the independent variable of the cepstrum and liftering for the operation of linear filtering the log magnitude spectrum by the operation of (5.5. Bogert et al.

e. k=0 Note the “tilde” ˜ symbol above x[n] in (5.58 Homomorphic Speech Analysis for use in processing sampled speech signals.3. several computational options exist.6) where x[n] is the complex cepstrum defined by (5. That is. Its ˆ purpose is to emphasize that using the DFT instead of the DTFT results in an approximation to (5. X[k] = X(ej2πk/N ) [89]). ˆ (5. Fortunately. This is Fig.5a) (5.3 Computing the cepstrum or complex cepstrum using the DFT.1b) due to the time-domain aliasing resulting from the sampling of the log of the DTFT [89].3. which shows that the complex cepstrum can be computed approximately using the equations N −1 X[k] = n=0 x[n]e−j(2πk/N )n (5.5c) and in Figure 5. An identical ˆ equation holds for the time-aliased cepstrum c[n].1b) as shown in Figure 5. 5.. A more serious problem in computation of the complex cepstrum is the computation of the complex logarithm.5b) (5. 5. . ∞ ˜ x[n] = ˆ r=−∞ x[n + rN ]. ˜ The effect of time-domain aliasing can be made negligible by using a large value for N .1 Computation Using the DFT Since the DFT (computed with an FFT algorithm) is a sampled (in frequency) version of the DTFT of a finite-length sequence (i.1a) and (5.5c) ˆ X[k] = log |X[k]| + j arg{X[k]} 1 ˜ ˆ x[n] = N N −1 ˆ X[k]ej(2πk/N )n . the DFT and inverse DFT can be substituted for the DTFT and its inverse in (5.1b).3.

7) Xmax (z) = z Mo k=1 (1 − ak z Xuc (z) = −1 )= k=1 (−ak ) k=1 (1 − a−1 z).5. are zeros of X(z) outside of the unit circle (|ak | > 1). 109. the phase of the sampled DTFT must be evaluated as samples of a continuous function of frequency. A variety of algorithms have been developed for phase unwrapping [90. since the phase is not used.. then it must be “unwrapped” modulo 2π in order to ensure that convolutions are transformed into additions. i.2 z-Transform Analysis The characteristic system for convolution can also be represented by the two-sided z-transform as depicted in Figure 5.e.3 Computation of the Cepstrum 59 because the angle of a complex number is usually specified modulo 2π. i. (1 − bk z −1 ) . phase unwrapping can be avoided by using numerical computation of the poles and zeros of the z-transform.. In order for the complex cepstrum to be evaluated properly.8c) (1 − ck z k=1 ) The zeros of Xmax (z). where Mo Mo Mo (5. If the principal value phase is first computed at discrete frequencies.8b) Muc (1 − ejθk z −1 ). Xmax (z) is thus the maximum-phase part of X(z).4.8a) k (5. it is not a problem in computing the cepstrum. . While accurate phase unwrapping presents a challenge in computing the complex cepstrum. (5. 127]. For this purpose. zk = ak . This is very useful for theoretical investigations. −1 k=1 Mi Xmin (z) = A k=1 N i (5.3. Furthermore. and recent developments in polynomial root finding [117] have made the z-transform representation a viable computational basis as well.e. we assume that the input signal x[n] has a rational z-transform of the form: X(z) = Xmax (z)Xuc (z)Xmin (z). 5. by the principal value.

(5.4 z-transform representation of characteristic system for convolution. This is .9) and collecting the coefficients of the positive and negative powers of z gives  Mo  an  k  n<0   n k=1    mo  (−ak ) n=0 x[n] = log |A| + log ˆ (5.11). The complex cepstrum of x[n] is determined by assuming that the complex logarithm log{X(z)} results in the sum of logarithms of each of the product terms. that are inside the unit circle (|bk | < 1 and |ck | < 1).. Mo Mo ˆ X(z) = log k=1 (−ak ) + k=1 Mi log(1 − a−1 z) + k Ni k=1 ∞ Muc k=1 log(1 − ejθk z −1 ) + log |A| + k=1 log(1 − bk z −1 ) − log(1 − ck z −1 ). 5.  n n n k=1 k=1 k=1 Given all the poles and zeros of a z-transform X(z). The factor z M o implies a shift of Mo samples to the left.11) allows us to compute the complex cepstrum with no approximation.e.9) Applying the power series expansion log(1 − a) = − n=1 an n |a| < 1 (5. It is included to simplify the results in (5. (5. respectively.10) to each of the terms in (5. i. Xuc (z) contains all the zeros (with angles θk ) on the unit circle.60 Homomorphic Speech Analysis Fig. where bk and ck are zeros and poles.11)   k=1    Mi n Ni n Muc jθk n   bk ck e   − − + n > 0. The minimum-phase part is Xmin (z).

This has become more feasible with increasing computational power and with new advances in finding roots of large polynomials [117]. An example would be the impulse response of an all-pole vocal tract model with system function H(z) = 1− k=1 G p = −k G p .11) is also useful as the basis for computation.. All that is needed is a process for obtaining the z-transform as a rational function and a process for finding the zeros of the numerator and denominator. From (5.3 Computation of the Cepstrum 61 the case in theoretical analysis where the poles and zeros are specified. −1 (5.. However.12) A second method that yields a z-transform is the method of linear predictive analysis.e. M X(z) = n=0 x[n]z −n = A Mo k=1 (1 − ak z −1 ) Mi k=1 (1 − bk z −1 ). (5. In this case. One method of obtaining a z-transform is simply to select a finitelength sequence of samples of a signal.e.13) αk z (1 − ck z k=1 ) Such models are implicit in the use of linear predictive analysis of speech (Chapter 6). 5. all the poles ck must be inside the unit circle for stability of the system.14) n > 0. The z-transform is then simply a polynomial with the samples x[n] as coefficients. . i. (5.3 Recursive Computation of the Complex Cepstrum Another approach to computing the complex cepstrum applies only to minimum-phase signals.11) it follows that the complex cepstrum of the impulse response h[n] corresponding to H(z) is  n<0 0    log |G| n = 0      p ˆ h[n] = k=1 cn k n (5. to be discussed in Chapter 6. i. signals having a z-transform whose poles and zeros are inside the unit circle.3.5.

5 shows an example of the short-time cepstrum of speech for the segments of voiced and unvoiced speech in Figures 4. working with the reciprocal (negative logarithm) of (5.17).16) h[n] = n−1  k ˆ  αn +  h[k]αn−k n > 0.5(a) and 4. p using (5. . it can be shown that there is a direct recursive relationship between the coefficients of the denominator polynomial in (5. (5. .4 Short-Time Homomorphic Filtering of Speech Figure 5.13). .  n<0 0    log G n=0 ˆ (5.  h[0] − n h[0] k=1 Furthermore. while the . The low quefrency part of the cepstrum is expected to be representative of the slow variations (with frequency) in the log spectrum. 1. it follows that p + 1 values of the complex cepstrum are sufficient to fully determine the speech model system in (5. 5.15) h[n] = ˆ  h[n] n−1 k h[k]h[n − k]    n ≥ 1.62 Homomorphic Speech Analysis It can be shown [90] that the impulse response and its complex cepstrum are related by the recursion formula:  n<0 0    log G n=0 ˆ (5..17) From (5.17). This fact is the basis for the use of the cepstrum in speech coding and speech recognition as a vector representation of the vocal tract properties of a frame of speech. i.13) and the complex cepstrum of the impulse response of the model filter.5(c).13) since all ˆ the denominator coefficients and G can be computed from h[n] for n = 0.16) it follows that the coefficients of the denominator polynomial can be obtained from the complex cepstrum through n−1 ˆ αn = h[n] − k=1 k ˆ h[k]αn−k n 1 ≤ n ≤ p.  n k=1 From (5. .e.

This periodic structure in the log spectrum of Figure 5. As a result. Note that the spectrum for the voiced segment in Figure 5.5.5(b) and 5. Furthermore. note that the rapid variations of the unvoiced spectra appear random with no periodic structure. high quefrency components would correspond to the more rapid fluctuations of the log spectrum. The existence of this peak in the quefrency range of expected pitch periods strongly signals voiced speech. . manifests itself in the cepstrum peak at a quefrency of about 9 ms in Figure 5.5(b). the quefrency of the peak is an accurate estimate of the pitch period during the corresponding speech interval. the autocorrelation function also displays an indication of periodicity. but not nearly as unambiguously as does the cepstrum. The log magnitudes of the corresponding short-time spectra are shown on the right as Figures 5.5(b).5.5 Short-time cepstra and corresponding STFTs and homomorphically-smoothed spectra.5(a). As shown in Figure 4. there is no strong peak indicating periodicity as in the voiced case.5(d). This is typical of Fourier transforms (periodograms) of short segments of random signals. This is illustrated by the plots in Figure 5.4 Short-Time Homomorphic Filtering of Speech (a) Voiced Cepstrum (b) Voiced log STFT 63 1 0 −1 0 10 20 30 40 4 2 0 −2 −4 −6 −8 0 2000 4000 6000 8000 (c) Unvoiced Cepstrum (d) Unvoiced log STFT 1 0 −1 0 10 20 time in ms 30 40 4 2 0 −2 −4 −6 −8 0 2000 4000 6000 frequency in kHz 8000 Fig. 5.5(b) has a structure of periodic ripples due to the harmonic structure of the quasi-periodic segment of voiced speech. On the other hand.

it is possible to separate information about the vocal tract contributions from the short-time speech spectrum [88].5(c)). which shows a plot that is very similar to the plot first published by Noll [83]. .5(d). For positions 8 through 15 the window only includes voiced speech.5 Application to Pitch Detection The cepstrum was first applied in speech processing to determine the excitation parameters for the discrete-time speech model of Figure 7. the quefrencies above 5 ms are multiplied by zero and the quefrencies below 5 ms are multiplied by 1 (with a short transition taper as shown in Figures 5. the window includes only unvoiced speech. Note again that the rapid variations of the unvoiced spectra appear random with no periodic structure.5(b) and 5. The successive spectra and cepstra are for 50 ms segments obtained by moving the window in steps of 12. From the discussion of Section 5.10. The DFT of the resulting modified cepstrum is plotted as the smooth curve that is superimposed on the short-time spectra in Figures 5. Therefore.64 Homomorphic Speech Analysis To illustrate the effect of liftering.6. and the quefrency of the peak is an accurate estimate of the pitch period during the corresponding speech interval. Noll [83] applied the short-time cepstrum to detect local periodicity (voiced speech) or the lack thereof (unvoiced speech). a useful perspective is that by liftering the cepstrum. On the other hand.5 ms (100 samples at a sampling rate of 8000 samples/sec). 5. respectively. while for positions 6 and 7 the signal within the window is partly voiced and partly unvoiced. it is apparent that for the positions 1 through 5. the spectra for voiced segments have a structure of periodic ripples due to the harmonic structure of the quasi-periodic segment of voiced speech. These slowly varying log spectra clearly retain the general spectral shape with peaks corresponding to the formant resonance structure for the segment of speech under analysis.4.5(a) and 5. On the left is a sequence of log short-time spectra (rapidly varying curves) and on the right is the corresponding sequence of cepstra computed from the log spectra on the left. the cepstrum peak at a quefrency of about 11–12 ms strongly signals voiced speech. This is illustrated in Figure 5. As can be seen from the plots on the right.

Presence of a strong peak implies voiced speech. As in most model-based signal processing applications of concepts such as the cepstrum.5. The essence of the pitch detection algorithm proposed by Noll is to compute a sequence of short-time cepstra and search each successive cepstrum for a peak in the quefrency region of the expected pitch period. for frames 6 and 7 the cepstrum peak is weak . 5. and the quefrency location of the peak gives the estimate of the pitch period.6 Short-time cepstra and corresponding STFTs and homomorphically-smoothed spectra. the pitch detection algorithm includes many features designed to handle cases that do not fit the underlying model very well.5 Application to Pitch Detection Short-Time Log Spectra in Cepstrum Analysis Short-Time Cepstra 65 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 window number 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 1 2 frequency in kHz 3 4 0 5 10 15 time in ms 20 25 Fig. For example.

5. As we have shown. the peak at twice the pitch period may be stronger than the peak at the quefrency of the pitch period. ω 2 (5. In such applications. a test pattern c[n] (vector of cepstrum values n = 1. we can assume that the cepstrum vector corresponds to a gain-normalized (c[0] = 0) minimumphase vocal tract impulse response that is defined by the complex cepstrum2 2c[n] 1 ≤ n ≤ nco ˆ h[n] = 0 n < 0.19b) ˆ where log |H(ej ω )| is the log magnitude of the DTFT of h[n] corresponding to the complex cepstrum in (5. (5. . a speech signal is represented on a frame-by-frame basis by a sequence of short-time cepstra. . ¯ For example. nco ) is compared against a similarly defined reference pattern c[n].19a) Equivalently in the frequency domain.6 Applications to Pattern Recognition Perhaps the most pervasive application of the cepstrum in speech processing is its use in pattern recognition systems such as design of vector quantizers (VQ) and automatic speech recognizers (ASR). the Euclidean distance applied to the cepstrum would give nco D= n=1 |c[n] − c[n]|2 . Noll applied temporal continuity constraints to prevent such errors. 2. . Since the ˆ cepstrum is always the even part of the complex cepstrum.18) or the real part of the 2 For minimum-phase signals. . . Such comparisons require a distance measure. the complex cepstrum satisfies h[n] = 0 for n < 0. ¯ (5. cepstra can be computed either by z-transform analysis or by DFT implementation of the characteristic system. it follows that h[n] = 2c[n] for n > 0. D= 1 2π π −π ˆ ¯ ˆ log |H(ej ω )| − log |H(ej ω )| dˆ .18) In problems such as VQ or ASR.66 Homomorphic Speech Analysis corresponding to the transition from unvoiced to voiced speech. In either case. In other problematic cases.

Therefore. i. then ∆cn [m] = ∆cn [m]. cepstrum-based comparisons are strongly related to comparisons of smoothed short-time spectra.20) where c(h) [m] will appear essentially the same in each frame.e. 3 In this section. 4 Stockham [124] showed how c(h) [m] for such linear distortions can be estimated from the signal y[n] by time averaging the log of the STFT. This property is extremely attractive in situations ˆ where the reference pattern c[m] has been obtained under different ¯ recording or transmission conditions from those used to acquire the test vector. the test vector can be compensated for the effects of the linear filtering prior to computing the distance measures used for comparison of patterns. where m denotes ˆ ˆ the quefrency index of the cepstrum. it will be useful to use somewhat more complicated notation. Therefore. if we can estimate c(h) [n].6. it can be removed by a simple first difference operation of the form: ∆cn [m] = cn [m] − cn−1 [m]. the short-time cepstrum of one frame of the filtered speech signal y[n] will be3 cn [m] = cn [m] + c(h) [m].e. the linear distortion effects ˆ ˆ ˆ are removed.6 Applications to Pattern Recognition 67 ˆ DTFT of h[n] in (5. If the analysis window is long compared to the length of h[n]. we can (x) (y) (x) obtain cn [m] at each frame from cn [m] by subtraction. ˆ ˆ (y) (x) (5. i. In these circumstances.4 which we assume is non-time-varying.5... cn [m] = ˆ ˆ (y) cn [m] − c(h) [m]. and its interpretation as a difference of log spectra suggests a match to auditory perception mechanisms. . (x) we denote the cepstrum at analysis time n of a signal x[n] as cn [m]. Thus. instead of x[n].18). Specifically. 5. the cepstrum offers an effective and flexible representation of speech for pattern recognition problems. Therefore. ˆ ˆ ˆ (y) (x) (y) (y) (y) (5.1 Compensation for Linear Filtering Suppose that we have only a linearly filtered version of the speech signal. Another approach to removing the effects of linear distortions is to observe that the cepstrum component due to the distortion is the same in each frame.21) It is clear that if cn [m] = cn [m] + c(h) [m] with c(h) [m] being indepenˆ ˆ (y) (x) dent of n. y[n] = h[n] ∗ x[n].

A solution to this problem is to use weighted distance measures of the form: nco D= n=1 g 2 [n] |c[n] − c[n]|2 .23) Tests of weighted distance measures showed consistent improvements in automatic speech recognition tasks. . . . it is observed that there is significant statistical variability due to a variety of factors including short-time analysis window position. Instead of g[n] = n for all n. . Itakura and Umezaki [55] used the group delay function to derive a different cepstrum weighting function. nco could be used to equalize the contributions for each term to the cepstrum difference. Juang et al. bias toward harmonic peaks. 126]. 2. ¯ (5. or . This suggests that g[n] = n for n = 1. and additive noise [63. . nco . 40] first noted that the sequence of cepstrum values has temporal information that could be of value for a speaker verification system. He used polynomial fits to cepstrum sequences to extract simple representations of the temporal variation. 5.22a) which can be written as the Euclidean distance of liftered cepstra nco D= n=1 |g[n]c[n] − g[n]¯[n]|2 . . (5. [63] observed that the variability due to the vagaries of LPC analysis could be lessened by using a lifter of the form: g[n] = 1 + 0. that when averaged over many frames of speech and speakers. 2. cepstrum values c[n] have zero means and variances on the order of 1/n2 .22b) Tohkura [126] found. .5nco sin(πn/nco ) n = 1. The delta cepstrum as defined in (5.6.68 Homomorphic Speech Analysis Furui [39. c (5. for example.2 Liftered Cepstrum Distance Measures In using linear predictive analysis (discussed in Chapter 6) to obtain cepstrum feature vectors for pattern recognition problems.21) is simply the slope of a first-order polynomial fit to the cepstrum time evolution. .

Note that the bandwidths in Figure 5.6 Applications to Pattern Recognition 69 the lifter of (5. This is significant in light of models for human perception of sound.24) This lifter has great flexibility. The basic idea is to compute a frequency analysis based upon a filter bank with approximately critical band spacing of the filters and bandwidths. in test conditions with additive white noise and also with linear filtering distortions. However. ˆ resulting in a DFT Xn [k] for analysis time n. weighted cepstrum distance measures have a directly equivalent interpretation in terms of distance in the frequency domain.6. as noted in Chapter 3. Itakura proposed the lifter g[n] = ns e−n 2 /2τ 2 .7. we have essentially g[n] = n for small n with high quefrency tapering. For 4 kHz bandwidth. 5. the difference in recognition rate was small for different values of s when τ ≈ 5 although performance suffered with increasing s for larger values of τ . a short-time Fourier analysis is done first. recognition rates improved significantly with τ = 5 and increasing values of the parameter s. With this in mind. This was attributed to the fact that for larger s the group delay spectrum becomes very sharply peaked and thus more sensitive to small differences in formant locations.7 are constant for center frequencies below 1 kHz and then increase exponentially up to half the sampling rate of 4 kHz resulting . if s = 0 we have simply lowpass liftering of the cepstrum.3 Mel-Frequency Cepstrum Coefficients As we have seen. which. For example. They found that for clean test utterances.23).5. In most implementations. Davis and Mermelstein [27] formulated a new type of cepstrum representation that has come to be widely used and known as the mel-frequency cepstrum coefficients (mfcc). approximately 20 filters are used. are based upon a frequency analysis performed in the inner ear. Then the DFT values ˆ are grouped together in critical bands and weighted by a triangular weighting function as depicted in Figure 5. Itakura and Umezaki [55] tested the group delay spectrum distance measure in an automatic speech recognition system. (5. If s = 1 and τ is large.

01 0. This normalization is built into the weighting functions of Figure 5.8 shows the result of mfcc analysis of a frame of voiced speech in comparison with the short-time Fourier spectrum. and a homomorphically smoothed spectrum. 5.7 Weighting functions for Mel-frequency filter bank. . R 2 (5. . Nmfcc . . in a total of 22 “filters.25b) is a normalizing factor for the rth mel-filter. 2. It is needed so that a perfectly flat input Fourier spectrum will produce a flat mel-spectrum.g. mfccn [m] is evaluated for a number of coefficients. . The large dots are the values of log (MFn [r]) and the line interpolated ˆ . i.25a) where Vr [k] is the triangular weighting function for the rth filter ranging from DFT index Lr to Ur . LPC spectrum (discussed in Chapter 6). Nmfcc = 13 and R = 22.e. For each frame. Figure 5. ˆ that is less than the number of mel-filters. e. ˆ k=Lr (5.005 0 0 1000 2000 frequency in Hz 3000 4000 Fig. ˆ mfccn [m] = ˆ 1 R R log (MFn [r]) cos ˆ r=1 1 2π r+ m . R as ˆ MFn [r] = ˆ 1 Ar Ur |Vr [k]Xn [k]|2 ..70 Homomorphic Speech Analysis 0.26) Typically.” The mel-frequency spectrum at analysis time n is defined for r = 1..7. where Ur Ar = k=Lr |Vr [k]|2 (5. a discrete cosine transform of the log of the magnitude of the filter outputs is computed to form the function mfccn [m].

between them is a spectrum reconstructed at the original DFT frequencies. It remains one of the most effective indicators of voice pitch that have been devised. At higher frequencies.5. Because the . 0 500 1000 1500 2000 2500 frequency in Hz 3000 3500 4000 Fig. Mel cepstrum smoothing. but they have in common that they have peaks at the formant resonances. Note that all these spectra are different. 5. the reconstructed mel-spectrum of course has more smoothing due to the structure of the filter bank. Note that the delta cepstrum idea expressed by (5. LPC smoothing. 5. the cepstrum entered the realm of speech processing as a basis for pitch detection.21) can be applied to mfcc to remove the effects of linear filtering as long as the frequency response of the distorting linear filter does not vary much across each of the mel-frequency bands.7 The Role of the Cepstrum 3 2 1 0 log magnitude 71 −1 −2 −3 −4 −5 −6 −7 Short-time Fourier Transform Homomorphic smoothing.7 The Role of the Cepstrum As discussed in this chapter.8 Comparison of spectral smoothing methods.

111]. Nevertheless. .72 Homomorphic Speech Analysis vocal tract and excitation components are well separated in the cepstrum. the linear predictive analysis methods to be discussed in the next chapter have proven to be more effective for a variety of reasons. 88. it was natural to consider analysis techniques for estimation of the vocal tract system as well [86. the cepstrum concept has demonstrated its value when applied to vocal tract system estimates obtained by linear predictive analysis. While separation techniques based on the cepstrum can be very effective.

The importance of this method lies both in its ability to provide accurate estimates of the speech parameters and in its relative speed of computation.1 Linear Prediction and the Speech Model We come to the idea of linear prediction of speech by recalling the source/system model that was introduced in Chapter 2. or random noise (during unvoiced speech). slowly time-varying system excited by either quasi-periodic impulses (during voiced speech). 6. In this chapter. The particular form of the source/system model implied by linear predictive analysis is depicted in Figure 6. Over short time intervals. we present a formulation of the ideas behind linear prediction. and discuss some of the issues that are involved in using it in practical speech applications. where the sampled speech signal was modeled as the output of a linear.1. the linear system is 73 . where the speech model is the part inside the dashed box.6 Linear Predictive Analysis Linear predictive analysis is one of the most powerful and widely used speech analysis techniques.

i. and the filter coefficients {ak } can be estimated in a very straightforward and computationally efficient manner by the method of linear predictive analysis.1 Model for linear predictive analysis of speech signals. The major advantage of this model is that the gain parameter. the speech samples s[n] are related to the excitation e[n] by the difference equation p s[n] = k=1 ak s[n − k] + Ge[n]. For the system of Figure 6. (6. (6. (6.3) and the prediction error. the excitation is whatever is needed to produce s[n] at the output of the system. G. αk .1) 1− k=1 ak z −k In linear predictive analysis.2) A linear predictor with prediction coefficients.4) . is p d[n] = s[n] − s[n] = s[n] − ˜ k=1 αk s[n − k].74 Linear Predictive Analysis Fig. the excitation is defined implicitly by the vocal tract system model. (6.1 with the vocal tract model of (6.. is defined as a system whose output is p s[n] = ˜ k=1 αk s[n − k].1). defined as the amount by which s[n] fails to ˜ exactly predict sample s[n].e. 6. described by an all-pole system function of the form: H(z) = S(z) = E(z) G p .

e.6) The basic problem of linear prediction analysis is to determine the set of predictor coefficients {αk } directly from the speech signal in order to obtain a useful estimate of the time-varying vocal tract system through the use of (6.4) that if the speech signal obeys the model of (6. That this approach will lead to useful results may not be immediately obvious. of (6. then it can be shown that the predictor coefficients that result from minimizing the mean-squared prediction error (over all time) are identical to the coefficients of (6. Thus.6. The basic approach is to find a set of predictor coefficients that will minimize the mean-squared prediction error over a short segment of the speech waveform. A second motivation for this approach follows from the fact that if a signal is generated by (6. the prediction error filter.2) and (6.4) it follows that the prediction error sequence is the output of an FIR linear system whose system function is p A(z) = 1 − k=1 αk z −k = D(z) . The resulting parameters are then assumed to be the parameters of the system function H(z) in the model for production of the given segment of the speech waveform.e.e. will be an inverse filter for the system. i.2). and if αk = ak .. recall that if αk = ak . Thus. A third pragmatic justification for using the minimum mean-squared prediction error as a . First. A(z) (6. P0 . This process is repeated periodically at a rate appropriate to track the phonetic variation of speech (i. i. but it can be justified in several ways. order of 50–100 times per second). A(z). then d[n] = Ge[n].. finding αk s that minimize the mean-squared prediction error seems consistent with this observation.2) with non-time-varying coefficients and excited either by a single impulse or by a stationary white noise input.1 Linear Prediction and the Speech Model 75 From (6.1). S(z) (6.6).. For voiced speech this means that d[n] would consist of a train of impulses. then d[n] = Ge[n].5) It can be seen by comparing (6. d[n] would be small except at isolated samples spaced by the current pitch period. H(z) = G .2) exactly. H(z).

8) That is.10) αk ϕn [i. k] = ϕn [i. . ˆ ˆ sn [m] = s[m + n] ˆ − M 1 ≤ m ≤ M2 .10) have a ˆ significant effect on the properties of the prediction coefficients that are obtained by solving (6.9) where the αk are the values of αk that minimize En in (6.11). .7) by setting ˆ ∂En /∂αi = 0. we will drop the tilde and use the notation αk to denote ˜ the values that minimize En . We can find the values of αk that minimize En in (6. ˆ ˆ 2 The quantities ϕ [i. . 0] i = 1. the solutions of (6.76 Linear Predictive Analysis basis for estimating the model parameters is that this approach leads to an exceedingly useful and accurate representation of the speech signal that can be obtained by efficient solution of a set of linear equations.7) where sn [m] is a segment of speech that has been selected in a neighˆ borhood of the analysis time n. . p.7). ˆ ˆ k=1 1 More (6. The notation ˆ denotes averaging over a finite number of samples. the time origin of the analysis segment is shifted to sample n of the entire signal. i. p..11) accurately. (Since the ˜ ˆ αk are unique. .) If we define ˆ ϕn [i. (6. k] = sn [m − i]sn [m − k] . 2.e. ˆ ˆ ˆ then (6. . The details of specific definitions of the averaging operation will be discussed below. thereby obtaining the equations1 ˆ p αk sn [m − i]sn [m − k] = sn [m − i]sn [m] ˜ ˆ ˆ ˆ ˆ k=1 1 ≤ i ≤ p. for i = 1. . k] are in the form of a correlation function for the speech segment n ˆ sn [m]. 2.7) only provide a stationary point which can be shown to be a minimum of En since En is a convex function of the αi s.9) can be written more compactly as2 p (6. . (6. The details of the definition of the averaging operation used in (6. The short-time average prediction error is defined as p 2 En = ˆ d2 [m] n ˆ = sn [m] − ˆ k=1 αk sn [m − k] ˆ . (6. .

e. in principle. ˆ To solve for the optimum predictor coefficients. k] when no confusion will result.13) Thus. 5] p En = ϕn [0.6. the optimum coefficients reduce En in (6. k]. We shall also find it advantageous to drop the subscripts n on En . k] and the subsequent soluˆ tion of the equations are somewhat intricate and further discussion is required.12) can be solved for the vector α = {αk } of unknown predictor coefficients that minimize the average squared prediction error for the segment sn [m]. ˆ ˆ ˆ ˆ 4 These two methods. 0] − ˆ ˆ k=1 αk ϕn [0.13) the most.9). We shall see below that two methods for linear predictive analysis emerge out of a consideration of the limits of summation and the definition of the waveform segment sn [m]. the details of the computation of ϕn [i. linear prediction analysis is very straightforward.7). we must first compute the quantities ϕn [i. As we have stated. Thus. the total minimum mean-squared error consists of a fixed component equal to the mean-squared value of the signal segment minus a term that depends on the predictor coefficients that satisfy (6.11). and ϕn [i. Once this ˆ is done we only have to solve (6. . k] for 1 ≤ i ≤ p and 0 ≤ k ≤ p. ˆ (6.4 ˆ 3 Although the αk s are functions of n (the time index at which they are estimated) this ˆ dependence will not be explicitly shown. i..3 Using (6.7) and (6. in a short-time analysis procedure. However. sn [m]. the averaging must be over a finite interval.2 Computing the Prediction Coefficients 77 If we know ϕn [i. applied to the same speech signal yield slightly different optimum predictor coefficients. this set of p equations ˆ in p unknowns. k] for 1 ≤ i ≤ p and 0 ≤ k ≤ p. 6. the minimum mean-squared prediction ˆ error can be shown to be [3.11) to obtain the αk s. which can be represented by the matrix equation: Φα = ψ. (6.2 Computing the Prediction Coefficients So far we have not been explicit about the meaning of the averaging notation used to define the mean-squared prediction error in (6.

k] needed in (6.15) ˆ Both (6. By changes of index of summation. M2 ϕn [i.11). Note the p “extra” samples ˆ (light shading) at the beginning that are needed to start the prediction error filter at time −M1 .14) where −M1 ≤ n ≤ M2 .14).2.16b) from which it follows that ϕn [i..11) inherit ˆ the same definition of the averaging operator. The third plot shows the prediction error computed with the optimum predictor coefficients.14) and (6. The quantities ϕn [i.e. i]. This method does not require any assumption about the signal outside the interval −M1 − p ≤ m ≤ M2 since the samples sn [m] for −M1 − p ≤ m ≤ M2 are sufficient to evaluate ˆ ϕ[i. (6. (6. Note that in solving for the optimum prediction coefficients using (6.78 Linear Predictive Analysis 6.15) can be expressed in the equivalent forms: M2 −i ϕn [i.16a) = m=−M1 −k sn [m]sn [m + k − i]. k] = ϕn [k. (6.2 shows the sequences that are involved in computing the mean-squared prediction error as defined by (6. The ˆ second plot shows that segment extracted as a finite-length sequence ˆ sn [m] = s[m + n] for −M1 − p ≤ m ≤ M2 . ˆ ˆ Figure 6. ˆ ˆ (6. k] for all required values of i and k. the . i. k] = ˆ m=−M1 −i M2 −k sn [m]sn [m + i − k] ˆ ˆ (6. The top part of this figure shows a sampled speech signal s[m]. k] = ˆ m=−M1 sn [m − i]sn [m − k] ˆ ˆ 1≤i≤p 0 ≤ k ≤ p.1 The Covariance Method One approach to computing the prediction coefficients is based on the definition M2 M2 p 2 En = ˆ m=−M1 (dn [m]) = ˆ m=−M1 2 sn [m] − ˆ k=1 αk sn [m − k] ˆ .15) require values of sn [m] = s[m + n] over the ˆ range −M1 − p ≤ m ≤ M2 . and the box denotes a segment of that waveform selected around some time index n.

6. With ˆ this method. .12) is a symmetric positive-semidefinite matrix.2 Illustration of windowing and prediction error for the covariance method.5) are not guaranteed to lie within the unit circle of the z-plane. In most cases of linear prediction. It is worthwhile to summarize them as follows: (C. The mathematical structure that defines the covariance method of linear predictive analysis implies a number of useful properties of the solution. it is theoretically possible for the average error to be exactly zero.1) The mean-squared prediction error satisfies En ≥ 0. the prediction error sequence is not explicitly computed since solution of (6.2 Computing the Prediction Coefficients 79 Fig. (C.3) The roots of the prediction error filter A(z) in (6. prediction error is implicitly computed over the range −M1 ≤ m ≤ M2 as required in (6. This implies that the vocal tract model filter (6.2) The matrix Φ in (6.15). 6.14) and (6.6) is not guaranteed to be stable. The minimum mean-squared prediction error would be simply the sum of squares of all the samples shown.11) does not require it. (C.

74.16b) leads to the conclusion that ∞ ϕn [i. the analysis segment sn [m] is defined as ˆ sn [m] = ˆ s[n + m]w[m] −M1 ≤ m ≤ M2 0 otherwise. ˆ ˆ ˆ (6.20) The resulting set of equations for the optimum predictor coefficients is therefore p αk φn [|i − k|] = φn [i] i = 1. Since the analysis segment is defined by the windowing of (6. ˆ m=−∞ (6. k] by φn [|i − k|]. Therefore. p. (6.17) where the analysis window w[m] is used to taper the edges of the segment to zero.2.16a) and (6.80 Linear Predictive Analysis (C.17) allows us to use the infinite limits to signify that the sum is over all nonzero values of dn [m]. 6. ˆ ˆ ˆ (6. 78].19) Thus. k] needed in (6. k] = ˆ m=−∞ sn [m]sn [m + |i − k|] = φn [|i − k|]. ϕ[i. . k] is a function only of |i − k|. Therefore.18) The windowing of (6. it follows that the prediction error sequence dn [m] can be nonzero only in the range ˆ −M1 ≤ m ≤ M2 + p. for example. the equations (6. ˆ ˆ k=1 (6.11) reduces to the STACF φn [|i − k|] that we disˆ ˆ cussed in Chapter 4 [53.11) can be solved efficiently using. En is defined as ˆ M2 +p ∞ En = ˆ m=−M1 (dn [m])2 = ˆ (dn [m])2 . we can replace ϕn [i. . .4) As a result of (C.21) . 54. the well known Cholesky decomposition of the covariance matrix Φ [3].2 The Autocorrelation Method Perhaps the most widely used method of linear predictive analysis is called the autocorrelation method because the covariance function ϕn [i. In the autocorrelation method. which is the STACF defined in Chapter 4 as ˆ ˆ ∞ φn [k] = ˆ m=−∞ sn [m]sn [m + k] = φn [−k]. Applying this notion ˆ to (6. 2.17) to be zero outside the interval −M1 ≤ m ≤ M2 . .2).

Also note the lightly shaded p samples at the beginning. Similarly.6.3 Illustration of windowing and prediction error for the autocorrelation method. These samples can be large due to the fact that the predictor must predict these samples from the zero-valued samples that precede the windowed segment sn [m]. It can also be seen that at least one of these last p samples of the prediction error must be nonzero.21)) is nonzero over the range −M1 ≤ m ≤ M2 + p. Figure 6. being the sum of ˆ . the last p samples of the prediction error can be large due to the fact that the predictor must predict zero-valued samples from windowed speech samples. It is easy to see that at least one of these first p samples ˆ of the prediction error must be nonzero. it follows that En . the prediction error (which is implicit in the solution of (6.3 shows the sequences that are involved in computing the optimum prediction coefficients using the autocorrelation method. Note that the zeroˆ valued samples outside the window are shown with light shading.2 Computing the Prediction Coefficients 81 Fig. Note that for this segment.2 with a Hamming window centered at time index n. The third plot shows the prediction error computed using the optimum coefficients. 6. The middle plot ˆ shows the result of multiplying the signal s[ˆ + m] by the window w[m] n and redefining the time origin to obtain sn [m]. The upper plot shows the same sampled speech signal s[m] as in Figure 6. For this reason.

a tapering window is generally used in the autocorrelation method.22) shows the detailed structure of the matrix equation Φα = ψ for the autocorrelation method.6) is guaranteed to be stable.      φ[1] φ[0] φ[1] · · · φ[p − 1] α1      φ[0] · · · φ[p − 2] α2  φ[2]  φ[1] (6. the mathematical structure of the autocorrelation method implies a number of properties of the solution.2). we discuss in more detail in Section 6.82 Linear Predictive Analysis squares of the prediction error samples.5 As in the case of the covariance method.3) The roots of the prediction error filter A(z) in (6. it is theoretically impossible for the error to be exactly zero because there will always be at least one sample at the beginning and one at the end of the prediction error sequence that will be nonzero.1) The mean-squared prediction error satisfies En > 0. (A.2) The matrix Φ in (6.3 The Levinson–Durbin Recursion As stated in (A. which means that all the elements on a given diagonal in the matrix are equal.3.12) is a symmetric positive-definite Toeplitz matrix [46].2) above. which. Equation (6. (A. With ˆ this method.12) is a symmetric positivedefinite Toeplitz matrix. (A. must always be strictly greater than zero. the matrix Φ in (6. . the equations (6. 6.4) As a result of (A.22)     =  ··· ··· · · ·  · · ·  · · ·   ··· αp φ[p] φ[p − 1] φ[p − 2] · · · φ[0] 5 For this reason. because of its many implications.5) are guaranteed to lie within the unit circle of the z-plane so that the vocal tract model filter of (6. including the following: (A.11) can be solved efficiently using the Levinson–Durbin algorithm.

3) (D. . . all predictors from order 0 (no prediction) to order p are computed along with the corresponding mean-squared prediction errors E (i) .23) leads (after some manipulations) to an interpretation of the Levinson–Durbin algorithm in terms of a lattice filter structure as in Figure 6.5) (D.4(a). . .4) can be used to show that the prediction error system function satisfies A(i) (z) = A(i−1) (z) − ki z −i A(i−1) (z −1 ). 2.3 The Levinson–Durbin Recursion 83 Note that the vector ψ is composed of almost the same autocorrelation values as comprise Φ. That algorithm. p αj = αj An important feature of the Levinson–Durbin algorithm is that it determines by recursion the optimum ith-order predictor from the optimum (i − 1)th-order predictor. . 2.6) j=1 = ki if i > 1 then for j = 1. known as the Levinson–Durbin algorithm.22) it is possible to derive a recursive algorithm for inverting the matrix Φ. .4) (D. .3) and (D. . . (6. .1)  αj (i−1) i−1 φ[i − j] E (i−1) (D.23) Defining the ith-order forward prediction error e(i) [n] as the output of the prediction error filter with system function A(i) (z) and b(i) [n] as the output of the ith-order backward prediction error filter B (i) (z) = z −i A(i) (z −1 ). is specified by the following steps: Levinson–Durbin Algorithm E0 = φ[0] for i = 1. Specifically. . the equations labeled (D.2) (D. . 2. and as part of the process.6. (6. . i − 1 (i) (i−1) (i−1) − ki αi−j αj = αj end 2 E (i) = (1 − ki )E (i−1) end (p) j = 1. Because of the special structure of (6. p  ki = φ[i] − (i) αi (D.

(a) Prediction error filter A(z).4(a) and work to the left eventually computing s[n] in terms of e[n]. In this case. Also. 6. . p play a key role in the Levinson–Durbin recursion and also in the lattice filter interpretation. The lattice structure corresponding to this is shown in Figure 6. 2. Lattice structures like this can be derived from acoustic principles applied to a physical model composed of concatenated lossless tubes [101].4 Lattice structures derived from the Levinson–Durbin recursion. They called the ki parameters. because they can be computed directly as a ratio of cross-correlation values between the forward and . showed that the parameters ki in the Levinson–Durbin recursion and the lattice filter interpretation obtained from it also could be derived by looking at linear predictive analysis from a statistical perspective.84 Linear Predictive Analysis Fig.6). 78. by solving two equations in two unknowns recursively. .4(b). the coefficients ki behave as reflection coefficients at the tube boundaries [3. PARCOR (for partial correlation) coefficients [54]. it is possible to start at the output of Figure 6. 101]. the sampled signals are related by a transfer function identical to (6. . . (b) Vocal tract filter H(z) = 1/A(z). Itakura and Saito [53. If the input and output signals in such physical models are sampled at just the right sampling rate. Note that the parameters ki for i = 1. 54].

4(a) removes part of the correlation in the input signal. Therefore. since E (i) = (1 − ki )E (i−1) is strictly greater than zero for predictors of all orders. It can be shown that this condition on the PARCORs also guarantees that all the zeros of a prediction error filter A(i) (z) of any order must be strictly inside the unit circle of the z-plane [74. φn [m]. from Equa2 tion (D. |Sn (ej ω )|2 .e. . 1.6). i. Specifically.24).4 LPC Spectrum 85 backward prediction errors at the output of the (i − 1)th stage of prediction in Figure 6. it must be true that −1 < ki < 1 for all i. 78]. Equation (D. p are used to compute the prediction coefficients and gain. the magnitude-squared of the frequency response of this system. each stage of Figure 6. which is the inverse discrete Fourier ˆ ˆ transform of the magnitude-squared of the STFT. obtained by evaluating H(z) on the unit circle at angles 2πf /fs . 6. it follows that. which in turn define the vocal tract system function H(z) in (6.. .5) of the algorithm.24) In the PARCOR interpretation. and the result is an algorithm for transforming the PARCOR representation into the linear predictor coefficient representation. (6. .6. ∞ e(i−1) [m]b(i−1) [m − 1] ki = m=−∞ ∞ e(i−1) [m] m=−∞ 2 ∞ b(i−1) [m − 1] m=−∞ 2 1/2 . . The PARCOR coefficients computed using (6. The Levinson–Durbin formulation provides one more piece of useful information about the PARCOR coefficients.4(a). The autocorrelation method is based on the short-time autocorrelation function.24) are identical to the ki obtained as a result of the Levinson–Durbin algorithm. Indeed.4 LPC Spectrum The frequency-domain interpretation of linear predictive analysis provides an informative link to our earlier discussions of the STFT and cepstrum analysis. .2) in the Levinson– Durbin algorithm can be replaced by (6. The values φn [m] for ˆ ˆ m = 0. of the ˆ windowed speech signal sn [m] = s[n + m]w[m].

Figure 6. 6. In (a) Voiced Autocorrelation (b) Voiced LPC Spectrum 1 0.5 0 −0.5 50 0 −50 −100 0 5 10 15 0 2000 4000 6000 8000 (c) Unvoiced Autocorrelation (d) Unvoiced LPC Spectrum 1 0.5(a) and 6. . These frequency responses are superimposed on the corresponding STFTs (shown in gray).86 Linear Predictive Analysis is of the form: 2 |H(ej2πf /fs )|2 = 1− G p .25) where the sampling frequency is fs = 16 kHz. The frequency responses of the corresponding vocal tract system models are computed using (6. and radiation.5 0 −0.5 shows a comparison between short-time Fourier analysis and linear predictive spectrum analysis for segments of voiced and unvoiced speech. Figures 6. the rapid variations with frequency in the STFT are due primarily to the excitation. with the first 23 values plotted with a heavy line. vocal tract transfer function. −j2πf /fs (6.5 50 0 −50 −100 0 5 10 time in ms 15 0 2000 4000 6000 frequency in kHz 8000 Fig.25) αk e k=1 and can be thought of as an alternative short-time spectral representation.5 Comparison of short-time Fourier analysis with linear predictive analysis. As we have observed before. while the overall shape is assumed to be determined by the effects of glottal pulse.5(c) show the STACF. These values are used to compute the predictor coefficients and gain for an LPC model with p = 22.

That this is true in general can be argued using the Parseval theorem of Fourier analysis [74].5 shows that a linear prediction model with p = 22 matches the general shape of the short-time spectrum. In linear predictive spectrum analysis. . the excitation effects are removed by focusing on the low-time autocorrelation coefficients. the excitation effects are removed by lowpass liftering the cepstrum. Evidently. The amount of smoothing of the spectrum is controlled by the choice of p.6 Effect of predictor order on: (a) estimating the spectral envelope of a 4-kHz bandwidth signal.6(a) are shown the STFT (in gray) and the frequency responses of a 12th-order model (heavy dark line) and a 40th-order model (thin dark line). In Figure 6. The question naturally arises as to how p should be chosen.6 offers a suggestion.4 LPC Spectrum 87 cepstrum analysis. This is in contrast to homomorphic Fig. and this is exactly what is desired.6. the linear predictive spectra tend to favor the peaks of the short-time Fourier transform. Figure 6. Figure 6. 6. but does not represent all its local peaks and valleys. and (b) the normalized mean-squared prediction error for the same signal.

Note the sharp decrease from p = 0 (no prediction implies V (0) = 1) to p = 1 and the less abrupt decrease thereafter. the LPC spectrum with p = 12. which tends toward an average of the peaks and valleys of the STFT. For example. In a particular application. a predictor order of p = 22 gave a good representation of the overall shape and resonance structure of the speech segment over the band from 0 to 8000 Hz.6(a). and the vocal tract imposes a resonance structure that. When coupled with an estimate of one resonance per kilohertz. Figure 6. the combination of the lowpass glottal pulse spectrum and the highpass filtering of radiation are usually adequately represented by one or two additional complex pole pairs. notice that the mean-squared error curve flattens out above about p = 12 and then decreases modestly thereafter. this leads to a rule of thumb of p = 4 + fs /1000. In this case.8. Observe that p = 40 gives a much different result.1) above. and radiation. see Figure 5. the vocal tract model is highly influenced by the pitch harmonics in the short-time spectrum. a homomorphically smoothed spectrum using 13 cepstrum values. It can be shown that if p is increased to the length of the windowed speech segment. The choice p = 12 gives a good match to the general shape of the STFT. for adult speakers. but as pointed out in ˆ (A. . the radiation filtering is highpass. vocal tract resonances. From the acoustic theory of speech production. highlighting the formant structure imposed by the vocal tract filter while ignoring the periodic pitch structure. it is common to use a predictor order of 10–12. Furthermore. E (2M ) does not go to zero [75]. Note that in Figure 6.5 where the sampling rate was fs = 16000 Hz. and a mel-frequency spectrum. that |H(ej2πf /fs )|2 → |Sn (ej2πf /fs )|2 . the prediction order is generally fixed at a value that captures the general spectral shape due to the glottal pulse. which shows a comparison between the short-time Fourier spectrum.88 Linear Predictive Analysis smoothing of the STFT. is comprised of about one resonance per kilohertz of frequency [101]. As an example. For the sampled speech signal.6(b) shows the normalized mean-squared prediction error V (p) = E (p) /φ[0] as a function of p for the segment of speech used to produce Figure 6. it follows that the glottal pulse spectrum is lowpass. for a sampling rate of fs = 8000 Hz.

In this section.7 shows an example of the roots (marked with ×) of a 12thorder predictor. These are the poles of H(z) that model the formant resonances. and the wide spacing between harmonics causes the peaks of the linear predictive vocal tract model to be biased toward those harmonics. (6. From these. but is less problematic for male voices where the spacing between harmonics is generally much smaller. if the model order is chosen judiciously as discussed in the previous section. 6. Therefore. In that example. we give only a brief summary of the most important equivalent representations. Note that eight (four complex conjugate pairs) of the roots are close to the unit circle. . Figure 6. particularly when they are used in speech coding. because we often wish to quantize the model parameters for efficiency in storage or transmission of coded speech.5. the zeros of A(z) are the poles of H(z).6 illustrates an important point about linear predictive analysis. which means that they only provide for the overall spectral shaping resulting from the glottal and radiation spectral shaping. This effect is a serious limitation for high-pitched voices. a variety of different equivalent representations can be obtained.5 Equivalent Representations 89 The example of Figure 6. 6.5) shows that the system function of the prediction error filter is a polynomial in z −1 and therefore it can be represented in terms of its zeros as p p A(z) = 1 − k=1 αk z −k = k=1 (1 − zk z −1 ). then it can be expected that roughly fs /1000 of the roots will be close in frequency (angle in the z-plane) to the formant frequencies. the speaker was a highpitched female.6).5 Equivalent Representations The basic parameters obtained by linear predictive analysis are the gain G and the prediction coefficients {αk }.26) According to (6. These different representations are important.1 Roots of Prediction Error System Function Equation (6.6. The remaining four roots lie well within the unit circle.

5.7 Poles of H(z) (zeros of A(z)) marked with × and LSP roots marked with ∗ and o.23). One possibility.27b) −(p+1) A(z −1 ). who defined the line spectrum pair (LSP) polynomials6 P (z) = A(z) + z −(p+1) A(z −1 ) Q(z) = A(z) − z 6 Note (6. 6.27a) (6.2 LSP Coefficients A much more desirable alternative to quantization of the roots of A(z) was introduced by Itakura [52].4(a) with an additional section with kp+1 = ∓1. would be to factor the polynomial as in (6. respectively. from which it follows that P (z) and Q(z) are system functions of lattice filters obtained by extending Figure 6. 6. It is well-known that the roots of a polynomial are highly sensitive to errors in its coefficients — all the roots being a function of all the coefficients [89]. it is important to maintain high accuracy in the location of the zeros of A(z). the similarity to (6. The prediction coefficients are perhaps the most susceptible to quantization errors of all the equivalent representations. Since the pole locations are crucial to accurate representation of the spectrum.26) and then quantize each root (magnitude and angle) individually. not often invoked.90 Linear Predictive Analysis Fig. .

Knowing the p LSFs that lie in the range 0 < ω < π is sufficient to completely define the LSP polynomials and therefore A(z). (LSP. the reconstructed polynomial A(z) will still have its zeros inside the unit circle. These properties make it possible to represent the linear predictor by quantized differences between the successive LSFs [119]. and Q(z) for a 12th-order predictor. i.7 and summarized below [52..e. which shows the roots of A(z). P (z). . (LSP.1) All the roots of P (z) and Q(z) are on the unit circle.28b) The normalized discrete-time frequencies (angles in the z-plane).e.3) is maintained through the quantization. An illustration of the LSP representation is given in Figure 6.4) The LSFs are close together when the roots of A(z) are close to the unit circle.. (LSP.5 Equivalent Representations 91 This transformation of the linear prediction parameters is invertible: to recover A(z) from P (z) and Q(z) simply add the two equations to obtain A(z) = (P (z) + Q(z))/2. 119]. i. Ωk and Θk .28a) Q(z) = (6.7.2) If p is an even integer. the LSFs are interlaced over the range of angles 0 to π.3) The PARCOR coefficients corresponding to A(z) satisfy |ki | < 1 if and only if the roots of P (z) and Q(z) alternate on the unit circle. (LSP. If property (LSP. are called the line spectrum frequencies or LSFs. then P (−1) = 0 and Q(1) = 0.6. k=1 (6. This new representation has some very interesting and useful properties that are confirmed by Figure 6. they can be represented as p+1 P (z) = k=1 p+1 (1 − ejΩk z −1 ) (1 − ejΘk z −1 ).

h[n]. since H(z) is a minimum-phase system (all its poles inside the unit circle). These relationships are particularly useful in speech recognition applications where a small number of cepstrum values is used as a feature vector. and (5. of the vocal tract filter. (6. Then.92 Linear Predictive Analysis 6.3 Cepstrum of Vocal Tract Impulse Response One of the most useful alternative representations of the linear predictor is the cepstrum of the impulse response. any desired number of cepstrum values can be computed. we can simply use them at step (D. which would require that the poles of H(z) be found by polynomial rooting. By working backward through the Levinson– Durbin algorithm. The impulse response can be computed recursively through the difference equation: p h[n] = k=1 αk h[n − k] + Gδ[n].16) gives the recursion for computing h[n] from the predictor coefficients and gain. This makes them an attractive parameter for quantization.3. we can compute the PARCOR coefficients from a given set of predictor coefficients. As discussed in Section 5.5. because of the minimum phase condition. However.4 PARCOR Coefficients We have seen that the PARCOR coefficients are bounded by ±1.5.17) gives the inverse recursion for computing the predictor coefficients from the complex cepstrum of the vocal tract impulse response. . 6.14).3. Equation (5. We have mentioned that if we have a set of PARCOR coefficients.29) or a closed form expression for the impulse response can be obtained by making a partial fraction expansion of the system function H(z).2) of the Levinson–Durbin algorithm thereby obtaining an algorithm for converting PARCORs to predictor coefficients. it follows that the complex cepstrum of h[n] can be computed using (5. a more direct recursive computation is posˆ sible [101]. The resulting algorithm is given below.

.5. . .5. . Ai 1 + ki (6. 78. .2) (P. i − 1 (i) (i) αj + ki αi−j (i−1) = αj 2 1 − ki end (i−1) ki−1 = αi−1 end 6.30) where Ai and Ai+1 are the areas of two successive tubes and the PARCOR coefficient −ki is the reflection coefficient for sound waves impinging on the junction between the two tubes [3. . .5 Equivalent Representations 93 Predictor-to-PARCOR Algorithm αj = αj j = 1. Such models are characterized by a set of tube cross-sectional areas denoted Ai . . . 2 for j = 1. 1 + egi (6. 101].4.5 Log Area Coefficients (p) (p) (P. Viswanathan and Makhoul [131] showed that the frequency response of a vocal tract filter represented by quantized log area ratio coefficients is relatively insensitive to quantization errors. These “area function” parameters are useful alternative representations of the vocal tract model obtained from linear predictive analysis. 2. The inverse transformation (from gi to ki ) is ki = 1 − egi .6. . Specifically. . p kp = αp for i = p.31) The PARCOR coefficients can be converted to predictor coefficients if required using the technique discussed in Section 6. .1) (P. p − 1.3) As mentioned above. the log area ratio parameters are defined as gi = log 1 − ki Ai+1 = log . the lattice filter representation of the vocal tract system is strongly related to concatenated acoustic tube models of the physics of sound propagation in the vocal tract. . 2.

The present chapter and Chapters 3–5 provide the basis for the remaining chapters of this text. and speech recognition. .94 Linear Predictive Analysis 6. which focus on these particular application areas. The value of linear predictive analysis stems from its ability to represent in a compact form the part of the speech model associated with the vocal tract. Over 40 years of research on linear predictive methods have yielded a wealth of knowledge that has been widely and effectively applied in almost every area of speech processing. we have discussed many of the fundamentals of linear predictive analysis of speech. which in turn is closely related to the phonemic representation of the speech signal. but especially in speech coding. speech synthesis.6 The Role of Linear Prediction In this chapter.

1.1. We begin by describing the basic operation of sampling a speech signal and directly quantizing and encoding of the samples. the first of three on digital speech processing applications.1 Sampling and Quantization of Speech (PCM) In any application of digital signal processing (DSP). 7. which comprise the process of A-to-D conversion. 95 . These operations. The remainder of the chapter discusses a wide variety of techniques that represent the speech signal in terms of parametric models of speech production and perception. focuses on specific techniques that are used in digital speech coding.7 Digital Speech Coding This chapter.1 1 Current A-to-D and D-to-A converters use oversampling techniques to implement the functions depicted in Figure 7. the first step is sampling and quantization of the resulting samples into digital form. are depicted for convenience in analysis and discussion in the block diagram of Figure 7.

1 The operations of sampling and quantization. 7.g. but never available for computations.2 depicts a typical quantizer definition. Theoretically. the example of Figure 7. the samples x[n] are real numbers that are useful for theoretical analysis. 4 kHz) and then use a sampling rate such as 8000 samples/s to avoid aliasing distortion [89]. The well known Shannon sampling theorem states that a bandlimited signal can be reconstructed exactly from samples taken at twice the highest frequency in the input signal spectrum (typically between 7 and 20 kHz for speech and audio signals) [89]. The quantizer simply takes the real number inputs x[n] and assigns an output x[n] accordˆ 2 In the case of ing to the nonlinear discrete-output mapping Q{ }. The sampler produces a sequence of numbers x[n] = xc (nT ). with samples within the peak-to-peak range being rounded and samples outside the range being “clipped” to either the maximum positive or negative level. we use a lowpass filter to remove spectral information above some frequency of interest (e. For samples within range. where T is the sampling period and fs = 1/T is the sampling frequency.2.1) that in this chapter.. the quantization error. not the ˆ complex cepstrum of x[n] as in Chapter 5. Figure 7. Often. the output samples are mapped to one of eight possible values. defined as e[n] = x[n] − x[n] ˆ 2 Note (7. . the notation x[n] denotes the quantized version of x[n].96 Digital Speech Coding A-to-D Converter xc (t ) Sampler x[n] Quantizer x[n] ˆ Q{ } Encoder c[n] T ∆ D-to-A Converter c′[n] Decoder x′[n] Reconstruction ˆ Filter xc (t ) ˆ′ ∆ Fig.

the step size will be ∆ = 2Xm /2B . satisfies the condition −∆/2 < e[n] ≤ ∆/2.2. (7.2 has 2B levels (1 bit usually signals the sign).1 represents the assigning of a binary code word to each quantization level. these code words are chosen to correspond to some convenient binary number system such that arithmetic can be done on the code words as . A B-bit quantizer such as the one shown in Figure 7. The block marked “encoder” in Figure 7. Because the quantization levels are uniformly spaced by ∆. 7. representation of a speech signal by binary-coded quantized samples is called pulse-code modulation (or just PCM) because binary numbers can be represented for transmission as on/off pulse amplitude modulation.2) where ∆ is the quantizer step size. such quantizers are called uniform quantizers. as in Figure 7. if the peak-to-peak range is 2Xm . These code words represent the quantized signal amplitudes.1 Sampling and Quantization of Speech (PCM) 97 Fig. and generally.7. Traditionally. Therefore.2 8-level mid-tread quantizer.

This means that we need a large peak-to-peak range so as to avoid clipping of the loudest sounds.e. the sampler. both these simple approaches degrade the perceived quality of the speech signal. The bit rate can be lowered by using fewer bits/sample and/or using a lower sampling rate. Note that the combination of the decoder and lowpass reconstruction filter represents the operation of D-to-A conversion [89]. (7. the maximum size of the quantization error is the same whether the signal sample is large or small. quantizer. 4 The denotes the possibility of errors in the codewords.4 Generally..1 or 48 kHz. for a given number of levels (bits). however. for a uniform quantizer. Furthermore. However. as the peak-to-peak range increases. the step size ∆ = 2Xm /2B is proportional to the peak-topeak range. but this is not the case if the step size is adapted from sample-to-sample. The operation of quantization confronts us with a dilemma. Therefore. the step size could be applied as depicted in the block labeled “Decoder” at the bottom of Figure 7.98 Digital Speech Coding if they were proportional to the signal samples.5 This value is more than adequate and much more than desired for most speech communication applications. i. have a wide dynamic range. instrumental music) are B = 16 and fs = 44. and coder are all integrated into one system. The data rate (measured in bits/second) of sampled and quantized speech signals is I = B · fs . The standard values for sampling and quantizing sound signals (speech.2 we can decrease the quantization error only by adding more levels (bits). . 5 Even higher rates (24 bits and 96 kHz sampling rate) are used for high-quality audio. This leads to a bitrate of I = 16 · 44100 = 705. speech especially. For a given peak-to-peak range for a quantizer such as the one in Figure 7. and only the binary code words c[n] are available. This chapter deals with a wide range of 3 In a real A-to-D converter.1. we do not need to be concerned with the fine distinction between the quantized samples and the coded samples when we have a fixed step size.2) states that the size of the maximum quantization error grows. their amplitudes vary greatly between voiced and unvoiced sounds and between speakers. singing. Symbol errors would cause additional error in the reconstructed samples. This is the fundamental problem of quantization. Most signals.3 In cases where accurate amplitude calibration is desired. 600 bits/s.

(Q. notably smooth input probability density functions and high rate quantizers. . e[n] acts like a white noise sequence. resulting 2 in average power σe = ∆2 /12.e.. i. With these assumptions. These simplifying assumptions allow a linear analysis that yields accurate results if the signal is not too coarsely quantized.1 Sampling and Quantization of Speech (PCM) 99 techniques for significantly reducing the bit rate while maintaining an adequate level of speech quality. but in this case the error is a deterministic function of the input and hence it cannot be independent of the input. These results can be easily shown to hold in the simple case of a uniformly distributed memoryless input and Bennett has shown how the result can be extended to inputs with smooth densities if the bit rate is assumed high [12]. Under these conditions.1 Uniform Quantization Noise Analysis A more quantitative description of the effect of quantization can be obtained using random signal analysis applied to the quantization error. nevertheless behaves as if it is a random signal with the following properties [12]: (Q. then the quantizer error will be uncorrelated with the quantizer output (however not the quantizer input. Bennett has shown that the correlations are only small because the error is small and.7. the noise samples appear to be uncorrelated from sample-tosample.1.3) The amplitudes of noise samples are uniformly distributed across the range −∆/2 < e[n] ≤ ∆/2. If the number of bits in the quantizer is reasonably high and no clipping occurs. under suitable conditions.2) Under certain assumptions. it is possible to derive the following formula for the signal-to-quantizing-noise ratio (in dB) of a B-bit 6 By “appear to be” we mean that measured correlations are small. (Q. as commonly stated). The condition “uncorrelated” often implies independence. it can be shown that if the output levels of the quantizer are optimized. the quantization error sequence. the correlations are equal to the negative of the error variance [12]. 7.1) The noise samples appear to be6 uncorrelated with the signal samples. although it is completely determined by the signal amplitudes.

The faint dashed lines are from (7. When σx gets close to Xm . and the assumptions underlying (7. respectively.3) where σx and σe are the rms values of the input signal and quantization noise samples. There is good agreement between these graphs indicating that (7. This accounts for the precipitous fall in SNR for 1 < Xm /σx < 8.3. 9. It can be way off. 10: Equation (7. 9. while σx depends on the input signal level. As signal level increases.3) is a reasonable estimate of SNR. Note that Xm is a fixed parameter of the quantizer.02B + 4.3) — light dashed lines. The formula of Equation (7. Comparison of -Law and Uniform Quantizers 60 50 40 SNR in dB 10 9 8 30 20 10 0 Fig.78 − 20 log10 Xm σx . the ratio Xm /σx decreases moving to the left in Figure 7. many samples are clipped.3) is an increasingly good approximation as the bit rate (or the number of quantization levels) gets large. . The measurements were done by quantizing 16-bit samples to 8. if the bit rate is not large [12]. Figure 7.3) no longer hold.3 Comparison of µ-law and linear quantization for B = 8.3) and the dark dashed lines are measured values for uniform quantization. Measured uniform quantization SNR — dark dashed lines. (7.100 Digital Speech Coding uniform quantizer: SNRQ = 10 log 2 σx 2 σe = 6. µ-law (µ = 100) compression — solid lines.3) with signal-to-quantizationnoise ratios measured for speech signals. however. 7. and 10 bits.3 shows a comparison of (7.

2 µ-Law Quantization Also note that the SNR curves in Figure 7. .7. increasing B by 1 bit (doubling the number of quantization levels) increases the SNR by 6 dB. µ-law quantization for speech is standardized in the CCITT G. but this is not possible because the log function is ill-behaved for small values of x[n]. In other words. defined as y[n] = Xm log 1 + µ |x[n]| Xm log(1 + µ) · sign(x[n]) (7. it should be noted that (7.1 Sampling and Quantization of Speech (PCM) 101 Also. µ = 255.3 show that the signal-to-noise ratio with µ-law compression remains relatively constant over a wide range of input levels.711 standard. If µ is large (> 100) this nonlinear transformation has the effect of distributing the effective quantization levels uniformly for small samples and logarithmically over the remaining range of the input samples. This is because the size of the quantization noise remains the same as the signal level decreases.1. This could be achieved if log(x[n]) were quantized instead of x[n].3 decrease linearly with increasing values of log10 (Xm /σx ).3 show that with all other parameters being fixed. it is exceedingly important to keep input signal levels as high as possible without clipping. log-PCM with fs = 8000 samples/s is widely used for digital telephony. 8-bit. µ-law compression (of the dynamic range) of the signal. On the other hand. it is also important to note that halving σx decreases the SNR by 6 dB.4) can be used prior to uniform quantization [118]. the SNR would be constant regardless of the size of the signal. cutting the signal level in half is like throwing away one bit (half of the levels) of the quantizer. 7. 8-bit log-PCM introduces little or no perceptible distortion. Quantization of µ-law compressed signals is often termed log-PCM. Once the bandwidth restriction has been imposed. This is an example of where speech quality is deliberately compromised in order to achieve a much lower bit rate.3) and Figure 7. As a compromise. In particular. Thus. The flat curves in Figure 7. If the quantization error were proportional to the signal amplitude.

” If the digital output of a codec is used as input to a speech processing algorithm. which are useful approximations to measured distributions for speech.4). Nowadays. It is based on the intuitive notion of constant percentage error. The fundamentals of optimum quantization were established by Lloyd [71] and Max [79]. 7.3 Non-Uniform and Adaptive Quantization µ-law compression is an example of non-uniform quantization. Lloyd [71] gave an algorithm for designing optimum non-uniform quantizers based on sampled speech signals. will incur less error than the least probable samples. Paez and Glisson [91] gave an algorithm for designing optimum quantizers for assumed Laplace and Gamma probability densities. it is necessary to know the probability distribution of the signal sample values so that the most probable samples. which for speech are the low amplitude samples. it is generally necessary to restore the linearity of the amplitudes through the inverse of (7. 7 Lloyd’s work was initially published in a Bell Laboratories Technical Note with portions of the material having been presented at the Institute of Mathematical Statistics Meeting in Atlantic City. To do this analytically. To apply this idea to the design of a non-uniform quantizer for speech requires an assumption of an analytical form for the probability distribution or some algorithmic approach based on measured distributions. readily available hardware devices convert analog signals directly into binary-coded µ-law samples and also expand compressed samples back to uniform scale for conversion back to analog signals. Such an A-to-D/D-to-A device is called a encoder/decoder or “codec. .7 Optimum non-uniform quantizers can improve the signal-to-noise ratio by as much as 6 dB over µ-law quantizers with the same number of bits.1. but with little or no improvement in perceived quality of reproduction. it was perceived to render speech equivalent to speech transmitted over the best long distance lines.102 Digital Speech Coding This configuration is often referred to as “toll quality” because when it was introduced at the beginning of the digital telephony era. New Jersey in September 1957. however. A more rigorous approach is to design a non-uniform quantizer that minimizes the mean-squared quantization error. Subsequently this pioneering work was published in the open literature in March 1982 [71].

Such adaption of the step size is equivalent to fixed quantization of the signal after division by the rms signal amplitude (En )1/2 .7. The adaptation speed can be adjusted by varying the analysis window size. Jayant [57] studied a class of feedback adaptive quantizers where the step size ∆[n] is a function of the step size for the previous sample. for example. To amortize this overhead. 58]. thereby adding to the information rate. if the step size control is based on past quantized samples. Adaptive quantizers can achieve about 6 dB improvement (equivalent to adding one bit) over fixed quantizers [57. and the step size information must be transmitted or stored along with the quantized sample. the step size is increased if the previous sample was quantized in one of the highest levels. ∆[n − 1]. 7. as in (4. it is not necessary to transmit the step size.2 Digital Speech Coding As we have seen. When the quantizer in a PCM system is adaptive. if the previous sample is quantized using step size ∆[n − 1] to one of the low quantization levels. the quantizer is a feedback adaptive quantizer. the quantizer is called a feed-forward adaptive quantizer. With this definition. the system is called an adaptive PCM (APCM) system.5) where ∆0 is a constant and En is the short-time energy of the speech ˆ signal defined. Virtually perfect perceptual quality is achievable with high data .2 Digital Speech Coding 103 Another way to deal with the wide dynamic range of speech is to let the quantizer step size vary with time. It can be derived at the decoder by the same algorithm as used for encoding. the step size will ˆ go up and down with the local rms amplitude of the speech signal. the step size could satisfy ∆[n] = ∆0 (En )1/2 . then the step size is decreased for the next sample. For example.6). On the other hand. feedforward quantization is usually applied to blocks of speech samples. On the other hand. ˆ (7. By basing the adaptation on the previous sample. If the step size control is based on the unquantized signal samples. the data rate of sampled and quantized speech is B · fs . In this approach.

Straightforward sampling and uniform quantization (PCM) is perhaps the only pure waveform coder. 8 If the signal that is reconstructed from the digital representation is not perceptibly different from the original analog speech signal. This work has led to a basic approach in speech coding that is based on the use of DSP techniques and the incorporation of knowledge of speech production and perception into the quantization process.8 By reducing B or fs . Most modern model-based coding methods are based on the notion that the speech signal can be represented in terms of an excitation signal and a time-varying vocal tract system. and reducing B too much will introduce audible distortion that may resemble random noise. There are many ways that this can be done.” . These two components are quantized.4 illustrates why linear predictive analysis can be fruitfully applied for model-based coding. Non-uniform quantization and adaptive quantization are simple attempts to build in properties of the speech signal (time-varying amplitude distribution) and speech perception (quantization noise is masked by loud sounds). The upper plot is a segment of a speech signal. and the latter model-based coders. Traditionally. Reducing fs requires bandwidth reduction.104 Digital Speech Coding rate. The main objective in digital speech coding (sometimes called speech compression) is to lower the bit rate while maintaining an adequate level of perceptual fidelity. Model-based coders are designed for obtaining efficient digital representations of speech and only speech. Since the 1930s. engineers and speech scientists have worked toward the ultimate goal of achieving more efficient representations. In addition to the two dimensions of quality and bit rate. the complexity (in terms of digital computation) is often of equal concern. and then a “quantized” speech signal can be reconstructed by exciting the quantized filter with the quantized excitation. the digital representation is often referred to as being “transparent. the bit rate can be reduced. The former are generally called waveform coders. speech coding methods have been classified according to whether they attempt to preserve the waveform of the speech signal or whether they only seek to maintain an acceptable level of perceptual quality and intelligibility. but perceptual quality may suffer. but these extensions still are aimed at preserving the waveform. Figure 7.

Overlap-add methods [89] can be used effectively for this purpose. Note that the prediction error sequence amplitude of Figure 7. One class of systems simply attempts to extract an excitation signal and a vocal tract system from the speech signal without any attempt to preserve a relationship between the waveforms of the original and the quantized speech. the line between waveform coders and model-based coders is not distinct.4 is a factor of five lower than that of the signal itself. A more useful classification of speech coders focusses on how the speech production and perception models are incorporated into the quantization process. If the quantized prediction error (residual) segment is used as input to the corresponding vocal tract filter H(z) = 1/A(z). Such systems are attractive because they can be implemented with modest computation. 9 To implement this time-varying inverse filtering/reconstruction requires care in fitting the segments of the signals together at the block boundaries. The lower plot is the output of a linear prediction error filter A(z) (with p = 12) that was derived from the given segment by the techniques of Chapter 6. 7. it is nevertheless suggestive of more practical schemes which we will discuss.2 0 100 200 sample index 300 400 Fig. an approximation to the original segment of speech would be obtained. which means that for a fixed number of bits.4 Speech signal and corresponding prediction error signal for a 12th-order predictor. this segment of the prediction error could be quantized with a smaller step size than the waveform itself.7. Today.2 Digital Speech Coding Speech Waveform 105 1 0 −1 0 100 200 300 400 Prediction Error Sequence 0. .9 While direct implementation of this idea does not lead to a practical method of speech coding.2 0 −0.

These systems are generally classified as waveform coders. but irrespective of the type of quantizer. Closed loop systems. Figure 7.1 Closed-Loop Coders Predictive Coding The essential features of predictive coding of speech were set forth in a classic paper by Atal and Schroeder [5]. Differential PCM systems are simple examples of this class. often do a good job of preserving the speech waveform while employing many of the same techniques used in open-loop systems. it can be uniform or non-uniform.3 7. 7. A second class of coders employs the source/system model for speech production inside a feedback loop. which we designate as open-loop coders. The quantizer Q can have a number of levels ranging from 2 to much higher. model-based systems that also preserve the waveform. Dudley early in the history of speech processing research [30]. but we prefer to emphasize that they are closed-loop. where e[n] is the quantization error. but increased availability of computational power has made it possible to implement much more sophisticated closed-loop systems called analysis-by-synthesis coders. which is comprised of a feedback structure that includes the blocks labeled Q and P . These compare the quantized output to the original input and attempt to minimize the difference between the two in some prescribed sense. although the basic principles of predictive quantization were introduced by Cutler [26]. so p x[n] = ˜ k=1 αk x[n − k]. The reader should ignore initially the blocks concerned with adaptation and all the dotted control lines and focus instead on the core DPCM system. and it can be adaptive or not. and thus are called closedloop coders.5 shows a general block diagram of a large class of speech coding systems that are called adaptive differential PCM (ADPCM) systems. ˆ the quantized output can be expressed as d[n] = d[n] + e[n].6) .3. are also called vocoders (voice coder ) since they are based on the principles established by H. The block labeled P is a linear predictor. ˆ (7.106 Digital Speech Coding These coders. since they explicitly attempt to minimize a time-domain distortion measure.

i. the predictor parameters are estimated from the input signal x[n]. Finally. the 10 In a feed-forward adaptive predictor. the difference between the input ˜ x[n].7.e.3 Closed-Loop Coders 107 Fig.5. they can also be estimated from past samples of the reconstructed signal x[n]. The latter would be termed a feedback adaptive ˆ predictor.. 7. ˆ and the predicted signal.5 General block diagram for adaptive differential PCM (ADPCM). the signal x[n] is predicted based on p past samples of the signal ˜ 10 The signal d[n] = x[n] − x[n]. . although. as shown in Figure 7. is the input to the quantizer.

but informative modification leads to SNR = 2 2 σx σd · 2 = GP · SNRQ . A simple. large number of quantization levels with small step size). the signal x[n] is the output of what we have called the “vocal ˆ ˆ tract filter. the signal-to-noise ratio of the ADPCM system is 2 2 SNR = σx /σe . It says that no matter what predictor is used in Figure 7. ˆ (7. SNRQ is the signal-to-noise ratio of the quantizer. step size information must be part of the digital representation.5.3). for fine quantization (i.. if GP > 1 it represents an improvement gained by placing a predictor based on the speech model inside the feedback loop around the quantizer. As indicated.108 Digital Speech Coding ˆ relationship between x[n] and d[n] is ˆ p x[n] = ˆ k=1 ˆ αk x[n − k] + d[n].e.5.” and d[n] is the quantized excitation signal. The feedback structure of Figure 7. then the variance of d[n] will be less than the variance of x[n] so it will be possible to use a smaller step size and therefore to reconstruct x[n] with less error than if x[n] were quanˆ tized directly. ˆ (7. is given by (7. From (7. Finally. If the prediction is good. The quantity GP is called the prediction gain.. i.e. which. 2 σd σe (7. as shown in Figure 7. As shown in Figure 7.5 contains within it a source/system speech model.7) i.or feedback-adaptive. since ˆ x[n] = x[n] + d[n] it follows that ˆ ˜ x[n] = x[n] + e[n]. Neglecting . the quantizer can be either feedforward. The number of bits B determines the bit-rate of the coded difference signal. the quantization error in x[n] is idenˆ ˆ tical to the quantization error in the quantized excitation signal d[n]. which.8).9) where the obvious definitions apply..8) This is the key result for DPCM systems.e.5(b). if the quantizer is feedforward adaptive. is the system needed to reconstruct the quantized speech signal x[n] from the ˆ coded difference signal input.

Fixed predictors can be designed based on long-term average correlation functions.6 shows an estimated long-term autocorrelation function for speech sampled at an 8 kHz sampling rate.10) as a function of predictor order p. αk ρ[k] (7.5 0 −0.5 0 2 4 6 autocorrelation lag 8 10 12 Fig. .6 Long-term autocorrelation function ρ[m] (lower plot) and corresponding prediction gain GP . 7. The predictor can be either fixed or adaptive.7.3 Closed-Loop Coders 109 the correlation between the signal and the quantization error. and adaptive predictors can be either feedforward or feedback (derived by analyzing past samples of the quantized speech signal reconstructed as part of the coder). The upper plot shows 10 log10 GP computed from (7. it can be shown that GP = 2 σx 2 = σd 1 p .10) 1− k=1 where ρ[k] = φ[k]/φ[0] is the normalized autocorrelation function used to compute the optimum predictor coefficients. Note that a first-order fixed predictor can yield about 6 dB prediction gain so that either the quantizer can have 1 bit less or the 10 G in dB 5 p 0 0 2 4 6 8 predictor order p 10 12 1 0. The lower figure in Figure 7.

The simplest delta modulators use a very high . studied. they nevertheless fit neatly into the general theory of predictive coding.5. adaptation algorithms are usually based on the bit stream at the output of the 1-bit quantizer.e. and not surprisingly. 1 − ρ2 [1] (7. To see how such systems can work. Great flexibility is inherent in the block diagram of Figure 7. Instead. and implemented as standard systems. and adapting the predictor at a phoneme rate can produce an additional 4 dB of gain [84]. Here we can only mention a few of the most important.3. 7. DM systems originated in the classic work of Cutler [26] and deJager [28].11) Furthermore.110 Digital Speech Coding reconstructed signal can have a 6 dB higher SNR. this can be compensated by the prediction gain due to oversampling.6. even though a 1-bit quantizer has a very low SNR. and in contrast to more general predictive coding systems. Higher-order fixed predictors can achieve about 4 dB more prediction gain. If the same bandwidth speech signal were over-sampled at a sampling rate higher than 8000 samples/s. Delta modulation systems can be implemented with very simple hardware.. many systems based upon the basic principle of differential quantization have been proposed. note that the optimum predictor coefficient for a first-order predictor is α1 = ρ[1]. both ρ[1] and α1 approach unity and GP gets large.2 Delta Modulation Delta modulation (DM) systems are the simplest differential coding systems since they use a 1-bit quantizer and usually only a first-order predictor. we could expect that the correlation value ρ[1] would lie on a smooth curve interpolating between samples 0 and 1 in Figure 7. i. so from (7. as fs gets large.10) it follows that the prediction gain for a first-order predictor is GP = 1 . This correlation function is for fs = 8000 samples/s. While DM systems evolved somewhat independently of the more general differential coding methods. Thus. note the nature of the long-term autocorrelation function in Figure 7. they are not generally implemented with block processing.6.

For these reasons.3. only bit synchronization is required in transmission. This limitation can be mitigated with only modest increase in complexity by using an adaptive 1-bit quantizer [47. 4. which introduces delay.5. Noll compared log PCM. he showed that doubling the sampling rate of his ADM system improved the SNR by about 10 dB. 56].5 are implemented on short blocks of input signal samples. and 5-bits with 8 kHz sampling rate. adaptive PCM. and three ADPCM systems all using quantizers of 2. Their bit rate.3.7. fixed DPCM.3 Adaptive Differential PCM Systems Differential quantization with a multi-bit quantizer is called differential PCM (DPCM). For example. .6 kHz. For all bit rates (16. the operations depicted in Figure 7.3. (3) Adding first-order fixed or adaptive prediction improved the SNR by about 4 dB over APCM. 24. See Section 7. must be very high to achieve good quality reproduction. being equal to fs . and it can be adjusted to be robust to channel errors. it has very low coding delay. A careful comparison of a variety of ADPCM systems by Noll [84] gives a valuable perspective on the relative contributions of the quantizer and predictor. adaptive differential PCM systems can have any combination of adaptive or fixed quantizers and/or predictors. 32. the results were as follows11 : (1) Log PCM had the lowest SNR. As depicted in Figure 7.3 Closed-Loop Coders 111 sampling rate with a fixed predictor. Furthermore. Jayant [56] showed that for bandlimited (3 kHz) speech. 11 The bit rates mentioned do not include overhead information for the quantized prediction coefficients and quantizer step sizes.1 for a discussion of this issue. adaptive delta modulation (ADM) is still used for terminal-cost-sensitive digital transmission applications. and 40 kbps). 7. adaptive delta modulation (ADM) with bit rate (sampling rate) 40 kbits/s has about the same SNR as 6-bit log PCM sampled at 6. (2) Adapting the quantizer (APCM) improved SNR by 6 dB. Generally. 3. Adaptive delta modulation has the following advantages: it is very simple to implement.

721. G. All these approaches raise the noise spectrum at low frequencies and lower it at high frequencies.4 for fs = 8 kHz work best. i. it can be filtered with the inverse system to restore the spectral balance. One of these concerns the quantization noise introduced by ADPCM.727). G. which we have seen is simply added to the input signal in the reconstruction process. the noise spectrum will take on the shape of the de-emphasis filter spectrum. and in the process. If γ is too close to one. filter the speech with a linear filter that boosts the high-frequency (HF) part of the spectrum.e. .112 Digital Speech Coding (4) A fourth-order adaptive predictor added about 4 dB. and increasing the predictor order to 12 added only 2 dB more. The classic paper by Atal and Schroeder [5] contained a number of ideas that have since been applied with great success in adaptive predictive coding of speech. which are often operated at 32 kbps where quality is superior to log-PCM (ITU G. The superior performance of ADPCM relative to log-PCM and its relatively low computational demands have led to several standard versions (ITU G.. A simple solution to this problem is to preemphasize the speech signal before coding.726.711) at 64 kbps. An equivalent approach is to define a new feedback coding system where the quantization noise is computed as the difference between the input and output of the quantizer and then shaped before adding to the prediction residual. thus taking advantage of the masking effects of the prominent low frequencies in speech [2]. it is clear that the noise will be most prominent in the speech at high frequencies where speech spectrum amplitudes are low. A simple pre-emphasis system of this type has system function (1 − γz −1 ) where γ is less than one. After reconstruction of the pre-emphasized speech.12 A more sophisticated application of this idea is to replace the simple fixed pre-emphasis filter with a time-varying filter designed to shape the quantization noise spectrum. If we invoke the white noise approximation for quantization error. Another idea that has been applied effectively in ADPCM systems as well as analysis-by-synthesis systems is to include a long-delay predictor to capture the periodicity as well as the short-delay correlation 12 Values around γ = 0. the low frequency noise is emphasized too much.

Even more complicated predictors with multiple delays and gains have been found to improve the performance of ADPCM systems significantly [2].3 Closed-Loop Coders 113 inherent in voiced speech.12) The parameter M is essentially the pitch period (in samples) of voiced speech. . and the parameter β accounts for amplitude variations between periods.5 shows that if the quantizer and predictor are fixed or feedback-adaptive.13 When this type of predictor is used.12). (7.6) to p x[n] = β x[n − M ] + ˜ ˆ k=1 αk (ˆ[n] − β x[n − M ]). Including the simplest long-delay predictor would change (7. The predictor parameters in (7. the “vocal tract system” used in the decoder would have system function     H(z) =    1 p 1− k=1 αk z   1   1 − βz −M −k  . 7. then only the coded difference signal is needed to reconstruct the speech signal (assuming that the decoder has the same coefficients and feedback-adaptation algorithms). the predictor coefficients and quantizer step size must be included as auxiliary data (side information) along with the quantized difference 13 Normally.3.1 Coding the ADPCM Parameters A glance at Figure 7. this analysis would be performed on the input speech signal with the resulting predictor coefficients applied to the reconstructed signal as in (7. x ˆ (7. Then the short-delay predictor parameters αk are estimated from the output of the long-delay prediction error filter. Otherwise.12) cannot be jointly optimized.13) Adding long-delay prediction takes more information out of the speech signal and encodes it in the predictor.3.7. One sub-optimal approach is to estimate β and M first by determining M that maximizes the correlation around the expected pitch period and setting β equal to the normalized correlation at M .

which requires 7 bits to cover the range of pitch periods to be expected with 8 kHz sampling rate of the input. almost all of which are preferable to direct quantization of the predictor coefficients themselves.. This gave a total of 20 bits/frame and added 2000 bps to the overall bit rate. To save bit rate. M + 1 and three gains each quantized to 4 or 5 bit accuracy. The number of bits per coefficient ranged from 5 for each of the lowest two PARCORs down to 1 each for the six highest-order PARCORs. M. 50–100 times/s. Vector Quantization for Predictor Quantization Another approach to quantizing the predictor parameters is called vector quantization. The difference signal will have the same sampling rate as the input signal. The total bit rate will be the sum of all the bit rates. If the parameters are up-dated 100 times/s. The basic principle of vector quantization is depicted in Figure 7. or VQ [45].g.7. The step size and predictor parameters will be estimated and changed at a much lower sampling rate.114 Digital Speech Coding signal. then a total of 4000 bps for the short-delay predictor information is required. As discussed in Chapter 6. The long-delay predictor in the system mentioned above used delays of M − 1. Each coefficient can be allocated a number of bits contingent on its importance in accurately representing the speech spectrum [131]. The PARCOR coefficients can be quantized with a non-uniform quantizer or transformed with an inverse sine or hyperbolic tangent function to flatten their statistical distribution and then quantized with a fixed uniform quantizer. e. yielding a total of 40 bits per frame. the short-delay predictors can be represented in many equivalent ways. Atal [2] reported a system which used a 20th-order predictor and quantized the resulting set of PARCOR coefficients after transformation with an inverse sine function. it is possible to up-date the short-delay predictor 50 times/s for a total bit rate contribution of 2000 bps [2]. Predictor Quantization The predictor parameters must be quantized for efficient digital representation of the speech signal. The long delay predictor has a delay parameter M . VQ can be applied in any context where a set of parameters naturally groups together into a vector (in the sense of .

14 For example the predictor coefficients can be represented by vectors of predictor coefficients. along with a distortion measure for comparing two vectors. L ≥ 10 · 2B .3 Closed-Loop Coders 115 Code Book 2B Vectors Code Book 2B Vectors vk ˆ v Distortion Measure Encoder Fig. the corresponding quantized vector vi can be looked up. . Typically a VQ training set consists of at least 10 times the ultimate number of codebook vectors. and it is facilitated by the LBG algorithm [70]. ˆ is returned from the exhaustive search. 2B codebook vectors are formed from the training set of vectors. The resulting codebook is then used to quantize general test vectors. vk ˆ i i Table Lookup Decoder vi ˆ linear algebra). or line spectrum frequencies.7 Vector quantization. i. a vector. to be quantized is compared exhaustively to a “codebook” populated with representative vectors of that type. log area ratios. The index i of the closest vector vi to v. One advantage of VQ is that the distortion measure can be designed to sort the vectors into perceptually relevant prototype vectors. The “training phase” is computationally intense. This index i then represents the quantized vector in the sense that if the codebook is known. Using an iterative process. 14 Independent quantization of individual speech samples or individual predictor coefficients is called scalar quantization. the index i can be represented by a B bit number. The design of a vector quantizer requires a training set consisting of a large number (L) of examples of vectors that are drawn from the same distribution as vectors to be quantized later. If the codebook ˆ has 2B entries. but need only be done once..7. v. In VQ. cepstrum values. 7.e. according to a prescribed distortion measure. PARCOR coefficients.

The blocks of zero samples can be encoded efficiently using variable-length-to-block codes yielding average bits/sample on the order of 0. 7.3. this coding method needs a total of 5600 bps for the difference signal. some sort of block coding must be applied. the speech signal must be bandlimited to somewhat less than 4 kHz by a lowpass filter. which is a compromise aimed at providing adequate intelligibility and moderate bit rate for voice communication. it is desirable to use as few bits as possible for the quantizer. Since the difference signal has the same sampling rate as the input signal. the additional bit rate would be B · fs .” which is a system that sets to zero value all samples whose amplitudes are within a threshold band and passes those samples above threshold unchanged.3.3.116 Digital Speech Coding The quantization process is also time consuming because a given vector must be systematically compared with all the codebook vectors to determine which one it is closest to. For high thresholds.7 [2]. For a sampling rate of 8 kHz. With this sampling rate. Adaptive prediction and quantization can easily lower the bit rate to 32 kbps with no degradation in perceived quality (with . One approach is to design a multi-bit quantizer that only operates on the largest samples of the difference signal. Difference Signal Quantization It is necessary to code the difference signal in ADPCM. This has led to numerous innovations in codebook structure that speed up the look up process [45]. the number of zero samples can be controlled. By adjusting the threshold. long runs of zero samples result. 7] proposed to precede the quantizer by a “center clipper. Atal and Schroeder [2. When this bit rate of the differential coder is added to the bit rate for quantizing the predictor information (≈4000 bps) the total comes to approximately 10. If straightforward B-bit uniform quantization is used. In order to reduce the bit rate for coding the difference samples. This was the approach used in earlier studies by Noll [84] as mentioned in Section 7.2 Quality vs.000 bps.3. Bit Rate for ADPCM Coders ADPCM systems normally operate with an 8 kHz sampling rate. and the entropy of the quantized difference signal becomes less than one.

this approach is clearly not optimal since the center clipper throws away information in order to obtain a sparse sequence.4 Analysis-by-Synthesis Coding While ADPCM coding can produce excellent results at moderate bit rates. raising the bit rate for ADPCM above 64 kbps cannot improve the quality over log-PCM because of the inherent frequency limitation. What is needed is a way of creating an excitation signal for the vocal tract filter that is both efficient to code and also produces decoded speech of high quality.3. 7.1 Basic Analysis-by-Synthesis Coding System Figure 7. x[n]. all the techniques discussed above and more must be brought into play. Of course.8 shows the block diagram of another class of closed-loop digital speech coders. between the input. This can be done within the same closed-loop framework as ADPCM. quality degrades significantly. The center clipping quantizer produces a difference signal that can be coded efficiently. its performance is fundamentally constrained by the fact that the difference signal has the same sampling rate as the input signal. the difference. These systems are called “analysis-by-synthesis coders” because the excitation is built up using an iterative process to produce a “synthetic” vocal tract filter output x[n] that matches the ˆ input speech signal according to a perceptually weighted error criterion. and where adequate transmission and/or storage capacity is available to support bit rates on the order of 10 kbps.7. the bit rate can be lowered below 32 kbps with only modest distortion until about 10 kbps. 7.3 Closed-Loop Coders 117 64 kbps log PCM toll quality as a reference). d[n]. As in the case of the most sophisticated ADPCM systems. but with some significant modifications. In particular. Thus. Below this value. x[n]. However. However. the operations of Figure 7. In order to achieve near-toll-quality at low bit rates.3. although not unreasonably so for current DSP hardware. This increases the system complexity. and the output of the vocal tract filter.8 are carried out on blocks of speech samples. is filtered with a linear filter called the ˆ . ADPCM is an attractive coding method when toll quality is required at modest cost.4.

it is desirable to shape its spectrum to take advantage of perceptual masking effects. In ADPCM. In ADPCM. In ADPCM.e. the input .8 Structure of analysis-by-synthesis speech coders.8 to the core ADPCM diagram of Figure 7. This is a key difference. x[n] = x[n] − d[n]. but instead of the synthetic output x[n].2 Perceptual Weighting of the Difference Signal Since −d[n] is the error in the reconstructed signal. W (z). the vocal tract model is in the same position in the closedloop system.” Note the similarity of Figure 7. the synthetic output differs from the input x[n] by the quantization error. As the first step in coding a block of speech samples. 7. a signal x[n] ˆ ˜ predicted from x[n] is subtracted from the input to form the difference ˆ signal. the reconstruction error is −d[n]. the excitation signal is determined from the perceptually weighted difference signal d [n] by an algorithm represented by the block labeled “Excitation Generator.118 Digital Speech Coding x[n] + - + d [n] Perceptual d′[n] Excitation Weighting W(z) Generator ˆ d [ n] x[n] ˆ Vocal Tract Filter H(z) Fig. and a percepˆ tually weighted version of that error is minimized in the mean-squared ˆ sense by the selection of the excitation d[n]. i.5. where an adaptive quantization algorithm operates on d[n] to produce a quanˆ tized difference signal d[n]. 7. perceptual weighting filter.4.. this is accomplished by quantization noise feedback or preemphasis/deemphasis filtering. Then. In analysis-by-synthesis. The perceptual weighting and excitation generator inside the dotted box play the role played by the quantizer in ADPCM. which is the input to the vocal tract system.3. both the vocal tract filter and the perceptual weighting filter are derived from a linear predictive analysis of the block. In analysis-by-synthesis coding.

. 30 20 log magnitude in dB 10 0 −10 −20 −30 0 1000 2000 frequency in Hz.3 Closed-Loop Coders 119 to the vocal tract filter is determined so as to minimize the perceptuallyweighted error d [n].9 and α2 = 0. If α1 > α2 the frequency response is like a “controlled” inverse filter for H(z). in this case. Figure 7.e. Thus.7. where typical values of α1 = 0. A(z/α2 ) H(z/α1 ) (7. 7. The weighting is implemented by linear filtering. 3000 4000 Fig.9 shows the frequency response of such a filter.14) the relative distribution of error can be adjusted. where. it follows that the error will be distributed in frequency so that relatively more error occurs at low frequencies.14).14) The poles of W (z) lie at the same angle but at α2 times the radii of the poles of H(z). d [n] = d[n] ∗ w[n] with the weighting filter usually defined in terms of the vocal tract filter as the linear system with system function W (z) = A(z/α1 ) H(z/α2 ) = . such errors would be masked by the high amplitude low frequencies. which is the shape desired.9 Comparison of frequency responses of vocal tract filter and perceptual weighting filter in analysis-by-synthesis coding. i. this filter tends to emphasize the high frequencies (where the vocal tract filter gain is low) and it deemphasizes the low frequencies in the error signal. Clearly. By varying α1 and α2 in (7. .4 are used in (7. and the zeros of W (z) are at the same angles but at radii of α1 times the radii of the poles of H(z).

.3 Generating the Excitation Signal Most analysis-by-synthesis systems generate the excitation from a finite fixed collection of input components.15) The βk s and the sequences fγk [n] are chosen to minimize L−1 E= n=0 ˆ ((x[n] − h[n] ∗ d[n]) ∗ w[n])2 .3. However. Since the component sequences are assumed to be known at both the coder and decoder.16).16) where h[n] is the vocal tract impulse response and w[n] is the impulse response of the perceptual weighting filter with system function (7. The error signal in the current frame at this initial stage of the iterative ˆ process would be d0 [n] = x[n] − x0 [n]. i. The weighted error at the first iteraˆ tion is d1 [n] = (d0 [n] − β1 fγ1 [n] ∗ h[n] ∗ w[n]).e. assume that we have determined which of the collection of input components fγ1 [n] will reduce the weighted mean-squared error the most and we have also determined the required gain β1 . the βs and γs are all that is needed to represent the ˆ input d[n]. Now at the first iteration stage. By superposition.. (7. ˆ x1 [n] = x0 [n] + β1 fγ1 [n] ∗ h[n].120 Digital Speech Coding 7. Normally ˆ this would be a decaying signal that could be truncated after L samples. and the perceptually weighted error would be d0 [n] = d0 [n] ∗ w[n]. the output in the current frame due to the excitation determined for previous frames is computed and denoted x0 [n] for 0 ≤ n ≤ L − 1. satisfactory results are obtained by solving for the component signals one at a time. N ˆ d[n] = k=1 βk fγk [n]. or if we define the perceptually weighted vocal tract impulse response as h [n] = h[n] ∗ w[n] 15 Several analysis frames are usually included in one linear predictive analysis frame. Starting with the assumption that the excitation signal is zero during the current excitation15 analysis frame (indexed 0 ≤ n ≤ L − 1). which we designate here as fγ [n] for 0 ≤ n ≤ L − 1. The input is composed of a finite sum of scaled components selected from the given collection. (7.4. It is very difficult to solve simultaneously for the optimum βs and γs that minimize (7. where L is the excitation frame length and γ ranges over the finite set of components.14).

(7. subtracting it from the previously computed residual error.18) is L−1 βk = dk−1 [n]yk [n] n=0 .20) While we have assumed in the above discussion that fγk [n] and βk are known. the corresponding equation takes the form: dk [n] = dk−1 [n] − βk yk [n]. The mean-squared error at stage k of the iteration is defined as L−1 L−1 Ek = n=0 (dk [n])2 = n=0 (dk−1 [n] − βk yk [n])2 . and so forth.18) where it is assumed that dk−1 [n] is the weighted difference signal that remains after k − 1 steps of the process. This process is continued by finding the next component.19) and the corresponding minimum mean-squared error is L−1 L−1 2 (dk−1 [n])2 − βk n=0 n=0 Ek (min) = (yk [n])2 .21) .17) where y [n] = fγk [n] ∗ h [n] is the output of the perceptually weighted vocal tract impulse response due to the input component fγk [n]. Generalizing to the kth iteration. Equation (7. Assuming that γk is known. L−1 2 (y [n]) n=0 (7.20) suggests that the mean-squared error can be minimized by maximizing L−1 L−1 2 βk n=0 2 (yk [n]) = 2 dk−1 [n]yk [n] n=0 L−1 (yk [n])2 n=0 . then the weighted error sequence is d1 [n] = d0 [n] − β1 fγ1 [n] ∗ h [n]. (7. (7. (7. we have not discussed how they can be found. the value of βk that minimizes Ek in (7.7.3 Closed-Loop Coders 121 and invoke the distributive property of convolution.

The error reduction process can be stopped (min) falls below a prescribed threshold.3 is quite general since the only constraint was that the collection of component input sequences be finite. 7.e. N need not be fixed in advance. i. This is because the components yk [n] are all just shifted versions of h [n].4.3. the corresponding βk can be found from (7. The excitation sequence derived by the process described in Section 7. the set of input sequences so determined can be assumed and a new set of βs can be found jointly by solving a set of N linear equations. an exhaustive search at each stage can be achieved by computing just one crosscorrelation function and locating the impulse at the maximum of the cross-correlation.17 In this case.16 If desired. and the new error sequence computed from (7.4 Multi-Pulse Excitation Linear Prediction (MPLP) The procedure described in Section 7. . In this system.22) where N is usually on the order of 4–5 impulses in a 5 ms (40 samples for fs = 8 kHz) excitation analysis frame. 16 Alternatively.3. (7. which normally is of 20 ms (160 samples) duration. fγ [n] = δ[n − γ].3 is therefore of the form: N ˆ d[n] = k=1 βk δ[n − γk ].19).17). The first analysis-by-synthesis coder was called multi-pulse linear predictive coding [4]. when Ek 17 It is advantageous to include two or more excitation analysis frames in one linear predictive analysis frame.. An exhaustive search through the collection of input components will determine which component fγk [n] will maximize the quantity in (7. the component input sequences are simply isolated unit impulse sequences.21) and thus reduce the mean-squared error the most.3.4.122 Digital Speech Coding which is essentially the normalized cross-correlation between the new input component and the weighted residual error dk−1 [n].4. the complete set of component input sequences and corresponding coefficients will have been determined. where γ is an integer such that 0 ≤ γ ≤ L − 1. Once this component is found. After N iterations of this process.

4. and the pulse amplitudes.3.3 Closed-Loop Coders 123 In a speech coder based on multi-pulse analysis-by-synthesis. only γ1 . it is not surprising that a great deal of effort has gone into schemes for finding excitation signals that are easier to encode than multipulse excitation. below about 10 kbps. the excitation signal components are Gaussian random sequences stored in a 18 Earlier work by Stewart [123] used a residual codebook (populated by the Lloyd algorithm [71]) with a low complexity trellis search. These usually involve some sort of differential coding scheme. The next major innovation in the history of analysis-by-synthesis systems was called code-excited linear predictive coding or CELP [112].5 Code-Excited Linear Prediction (CELP) Since the major portion of the bit rate of an analysis-by-synthesis coder lies in the excitation signal. but yet maintain high quality of reproduction of the speech signal. Specifically after determining the location γ1 of the first impulse. RPE requires significantly less computation than full-blown multipulse excitation analysis. the speech signal is represented by quantized versions of the prediction coefficients. the integer multiples. One version of RPE is the basis for the 13 kbps digital coder in the GSM mobile communications system. and the gain constants must be encoded. Thus.18 In this scheme.7. quality degrades rapidly due to the many parameters that must be coded [13]. however. Because the impulse locations are fixed after the first iteration. At bit rates on the order of 10–16 kbps multipulse coding can approach toll quality. the pulse locations. all other impulses are located at integer multiples of a fixed spacing on either side of γ1 . Using a pre-set spacing of 4 and an excitation analysis window of 40 samples yields high quality at about 10 kbps [67]. . Regular-pulse excitation (RPE) is a special case of multipulse excitation where it is easier to encode the impulse locations. 7. The predictor coefficients can be coded as discussed for ADPCM by encoding any of the alternative representations either by scalar or vector quantization. A number of different approaches have been devised for coding the excitation parameters.

CELP coders can introduce more than 40 ms delay into communication systems. remember that. If N sequences are used to form the excitation. the object of linear prediction is to produce just such a signal. The ITU G. then a total of M · N bits will be required to code the sequences. then a given sequence can be specified by an M -bit number. This is because each codebook sequence must be filtered with the perceptually weighted impulse response before computing the cross correlation with the residual error sequence. Efficient searching schemes and structured codebooks have eased the computational burden. 512 or 1024. The GSM half rate coder operating at 5600 bps is based on the IS-54 VSELP coder.”19 If the codebook contains 2M sequences. With this method.124 Digital Speech Coding “codebook. Another system called VSELP (vector sum excited linear prediction). The Federal Standard 1016 (FS1016) was adopted by the Department of Defense for use in secure voice transmission at 4800 bps. and modern DSP hardware can easily implement the computations in real time. was adopted in 1989 for North American Cellular Systems.729 conjugate-structure algebraic CELP (CS-ACELP) coder that is used in some mobile communication systems. in fact. Typical codebook sizes are 256. The main disadvantage of CELP is the high computational cost of exhaustively searching the codebook at each stage of the error minimization. the sequence lengths are typically L = 40–60 samples (5–7. 19 While it may seem counter-intuitive that the excitation could be comprised of white noise sequence.5 ms at 8 kHz sampling rate). which uses multiple codebooks to achieve a bit rate of 8000 bps.728 low-delay CELP coder uses very short excitation analysis frames and backward linear prediction to achieve high quality at a bit rate of 16. . Due to the block processing structure. the decoder must also have a copy of the analysis codebook so that the excitation can be regenerated at the decoder. Within the codebook. Still another standardized CELP system is the ITU G. Additional bits are still required to code the βk s. The basic CELP framework has been applied in the development of numerous speech coders that operate with bit rates in the range from 4800 to 16000 bps.000 bps with delays less than 2 ms.

If we maintain a memory of one or more frames of ˆ previous values of the excitation signal. and CELP coding has not been used at bit rates below about 4800 bits/s. For bit rates below 4800 bps. 7. engineers have turned to the vocoder principles that were established decades ago. We call these systems open-loop systems because they do not determine the excitation by a feedback process.4 Open-Loop Coders ADPCM coding has not been useful below about 9600 bps. it has not been possible to generate excitation signals that can be coded at low bit rates and also produce high quality synthetic speech.10 shows a source/system model for speech that is closely related to physical models for speech production.6 Long-Delay Predictors in Analysis-by-Synthesis Coders Long-delay predictors have an interesting interpretation in analysis-bysynthesis coding. Then the ˆ component β0 d[n − γ0 ] can be subtracted from the initial error to start the iteration process in either the multipulse or the CELP framework. 7.1 The Two-State Excitation Model Figure 7.4.3. The first step of the excitation computation would be to compute γ0 using (7.4.19). multipulse coding has similar limitations. The long segment of the past excitation acts as a sort of codebook where the L-sample codebook sequences overlap by L − 1 samples. This incorporation of components of the past history of the excitation has been referred to as “self-excitation” [104] or the use of an “adaptive codebook” [19]. As we have discussed .4 Open-Loop Coders 125 7. we can add a term β0 d[n − γ0 ] to (7.15).21) and then β0 using (7. this requires that the weighted vocal tract impulse response be convolved with L-sample segments of the past excitation starting with sample γ0 . This can be done recursively to save computation.7. and long-delay predictors are used in many of the standardized coders mentioned above. The additional bit rate for coding β0 and γ0 is often well worthwhile. For each value of γ0 to be tested. While these closed-loop systems have many attractive features.

1 Pitch. 7. where the spacing between impulses is the pitch period. with time. and voiced sounds are produced by a periodic impulse train excitation. the lowpass frequency shaping of the glottal pulse. as discussed in Chapter 8. The fundamental frequency varies slowly. and the parameters of the linear system are supplied at periodic intervals (frames) then the model becomes a speech synthesizer. The V/UV (voiced/unvoiced excitation) switch produces the alternating voiced and unvoiced segments of speech. Gain. radiation at the lips.10 Two-state-excitation model for speech synthesis. When the parameters of the model are estimated directly from a speech signal. the excitation model can be very simple. P0 . and the gain parameter G controls the level of the filter output. in the previous chapters.1. as we prefer. 7. P0 . in the case of voiced speech. and. the combination of estimator and synthesizer becomes a vocoder or. Unvoiced sounds are produced by exciting the system with white noise. more or less at the same rate as the vocal tract motions. . and V/UV Detection The fundamental frequency of voiced speech can range from well below 100 Hz for low-pitched male speakers to over 250 Hz for high-pitched voices of women and children. The slowly-time-varying linear system models the combined effects of vocal tract transmission. It is common to estimate the fundamental frequency F0 or equivalently. an open-loop analysis/synthesis speech coder. We shall refer to this as the two-state excitation model. When values for V/UV decision. G.4.126 Digital Speech Coding Fig.

For this purpose. the autocorrelation function value φn [0] at lag 0 or ˆ the cepstrum value cn [0] at quefrency 0 can be used to determine the ˆ energy of the segment of the input signal. much lower total bit . short segments of speech are analyzed to detect periodicity (signaling voiced speech) or aperiodicity (signaling unvoiced speech).8. In this case. The time variation of the pitch period is evident in the upper plots. which shows autocorrelation functions for both voiced speech. Since the vocal tract filter can be coded as in ADPCM or CELP. this totals 600 bps. It should be chosen so that the short-time energy of the synthetic output matches the short-time energy of the input speech signal. it is common to preprocess the speech by a spectrum flattening operation such as center clipping [96] or inverse filtering [77]. Typical values are 7 bits for pitch period (P0 = 0 signals UV) and 5 bits for G. and gain must be quantized. To do this. the pitch period. The STACF is also useful for this purpose as illustrated in Figure 4. The gain parameter G is also found by analysis of short segments of speech. which shows no evidence of periodicity. In pitch detection applications of the short-time autocorrelation. and measuring the times between the peaks [43]. and unvoiced speech.7. Another approach to pitch detection is suggested by Figure 5. For a frame rate of 50 frames/s. This preprocessing tends to enhance the peak at the pitch period for voiced speech while suppressing the local correlation due to formant resonances. a strong peak in the expected pitch period range signals voiced speech. Similarly. One of the simplest. and the location of the peak is the pitch period. which is well below the bit rate used to encode the excitation signal in closed-loop coders such as ADPCM or CELP. Figure 5. evincing a peak at the pitch period. yet most effective approaches to pitch detection. V/UV. lack of a peak in the expected range signals unvoiced speech [83]. operates directly on the time waveform by locating corresponding peaks and valleys in the waveform. which shows the cepstra of segments of voiced and unvoiced speech.6 shows a sequence of short-time cepstra which moves from unvoiced to voiced speech in going from the bottom to the top of the figure.4 Open-Loop Coders 127 the pitch period P0 = 1/F0 at a frame rate of about 50–100 times/s.5. For digital coding applications.

1.2 Vocal Tract System Estimation The vocal tract system in the synthesizer of Figure 7. s[n] = Ge[n] ∗ h[n]. the impulse response was digitally coded by quantizing each cepstrum value individually (scalar quantization) [86]. and the resulting output can overlap into the next frame. This comes at a large cost in quality of the synthetic speech output.10 can take many forms. For example. In the original homomorphic vocoder. respectively. The impulse response. reconstructed from the quantized cepstrum at the synthesizer. Such a speech coder is called a formant vocoder. homomorphic filtering was used to remove excitation effects in the short-time spectrum and then three formant frequencies were estimated from the smoothed spectra. voicing. and gain information.128 Digital Speech Coding rates are common in open-loop systems.3. is simply convolved with the excitation created from the quantized pitch. however. Figure 5.20 In a more recent application of homomorphic analysis in an analysis-by-synthesis framework [21]. one cepstrum computation can yield both an estimate of pitch and the vocal tract impulse response. In still another approach to digital coding. Thus.4. 7.e.3.. the cepstrum values were coded using vector quantization. . The Homomorphic Vocoder Homomorphic filtering can be used to extract a sequence of impulse responses from the sequence of cepstra that result from short-time cepstrum analysis. and the excitation derived by analysis-by-synthesis as described in Section 7.4. LPC Vocoder Linear predictive analysis can also be used to estimate the vocal tract system for an open-loop coder with two-state 20 Care must be taken at frame boundaries. The primary methods that have been used have been homomorphic filtering and linear predictive analysis as discussed in Chapters 5 and 6. These formant frequencies were used to control the resonance frequencies of a synthesizer comprised of a cascade of second-order section IIR digital filters [111].6 shows an example of the formants estimated for voiced speech. the impulse response can be changed at the time a new pitch impulse occurs. i.

voicing and gain and the remaining bits allocated to the vocal tract filter coded as PARCOR coefficients. This figure shows an example of inverse filtering of the speech signal using a prediction error filter. The two-state model attempts to construct the excitation signal by direct analysis and measurement of the input speech signal. which is quite similar to voice-excited vocoders (VEV) [113. and coding the samples with adaptive quantization.7. The speech signal can be reconstructed from the residual by passing it through the vocal tract system H(z) = 1/A(z). This system is also called LPC-10 (or LPC-10e) because a 10th-order covariance linear predictive analysis is used to estimate the vocal tract system. Adaptive delta modulation was used in [128]. lowering the sampling rate.11 shows a block diagram of a RELP coder [128]. and accurate coding could require several bits per sample.2 Residual-Excited Linear Predictive Coding In Section 7. Systems that attempt to code the residual signal directly are called residual-excited linear predictive (RELP) coders. A vocoder of this type was standardized by the Department of Defense as the Federal Standard FS1015.4 Open-Loop Coders 129 excitation [6]. Direct coding of the residual faces the same problem faced in ADPCM or CELP: the sampling rate is the same as that of the input. 7. Figure 7. the problem of reducing the bit rate of the residual signal is attacked by reducing its bandwidth to about 800 Hz. The LPC-10 system has a bit rate of 2400 bps using a frame size of 22. In this system.4. ADPCM and analysis-by-synthesis systems derive the excitation to the synthesis filter by a feedback process. 130].5 ms with 12 bits/frame allocated to pitch. In this case. but APCM could be used if the sampling rate is . the prediction coefficients can be coded in one of the many ways that we have already discussed.4 as motivation for the use of the source/system model in digital speech coding.2. we presented Figure 7. where the prediction error (or residual) is significantly smaller and less lowpass in nature. None of the methods discussed so far attempt to directly code the prediction error signal in an open-loop manner. Such analysis/synthesis coders are called LPC vocoders.

130 Digital Speech Coding x[n] Inverse Filter A(z) Linear Predictive Analysis r[n] Lowpass & Decimator Adaptive Quantizer d [n] {α } Encoder Interpolator ˆ[ n] Synthesis & Spectrum Filter Flattener H(z) d [ n] Adaptive Quantizer Decoder x[n] {α } Decoder Fig. The reduced bandwidth residual is restored to full bandwidth prior to its use as an excitation signal by a nonlinear spectrum flattening operation. The output of such systems is often described as “buzzy. White noise is also added according to an empirically derived recipe. the sampling rate of the input was 6.8 kHz and the total bit rate was 9600 kbps with 6800 bps devoted to the residual signal. errors in estimating pitch period or voicing decision cause the speech to sound unnatural if not unintelligible. The weaknesses of the two-state model .11 Residual-excited linear predictive (RELP) coder and decoder. its basic principles can be found in subsequent open-loop coders that have produced much better speech quality at bit rates around 2400 bps. While this system did not become widely used. lowered to 1600 Hz. the quality of the synthetic speech output leaves much to be desired. 7. and no pitch detection is required. 7. The 800 Hz band is wide enough to contain several harmonics of the highest pitched voices.4. In the implementation of [128].3 Mixed Excitation Systems While two-state excitation allows the bit rate to be quite low.” and in many cases. The quality achieved at this rate was not significantly better than the LPC vocoder with a two-state excitation model. which restores higher harmonics of voiced speech. The principle advantage of this system is that no hard V/UV decision must be made.

22 and a spectrally flat “pulse dispersion filter” whose Fig.12 Mixed-excitation linear predictive (MELP) decoder. . which focused one-by-one on the sources of distortions manifest in the two-state excitation coder such as buzziness and tonal distortions. a “jitter” parameter ∆P is invoked to better model voicing transitions. they each pass through a multi-band spectral shaping filter. The gains in each of five bands are coordinated between the two filters so that the spectrum of e[n] is flat.4 Open-Loop Coders 131 for excitation spurred interest in a mixed excitation model where a hard decision between V and UV is not required. This configuration was developed as the result of careful experimentation. [75] and greatly refined by McCree and Barnwell [80].” This represents adaptive spectrum enhancement filters used to enhance formant regions. The main feature is that impulse train excitation and noise excitation are added instead of switched. filters are also used routinely in CELP coders and are often referred to as post filters. Such a model was first proposed by Makhoul et al.12 depicts the essential features of the mixed-excitation linear predictive coder (MELP) proposed by McCree and Barnwell [80]. Other important features of the MELP system are lumped into the block labeled “Enhancing Filters. Figure 7. 21 Devoicing 22 Such is evident in frame 9 in Figure 5. This mixed excitation helps to model short-time spectral effects such as “devoicing” of certain bands during voiced speech.21 In some situations.7. 7.6. Prior to their addition.

5 Frequency-Domain Coders Although we have discussed a wide variety of digital speech coding systems. the output x[n] is not equal to the input. Each of these is followed by a down-sampler by a factor appropriate for the bandwidth of its bandpass input. Several new parameters of the excitation must be estimated at analysis time and coded for transmission. In fact. There is much room for variation within the general frameworks that we have identified. then it is possible for the output x[n] to be virtually identical to the input x[n] [24]. which operates at 2400. There is one more class of coders that should be discussed.. The MELP coder is said to produce speech quality at 2400 bps that is comparable to CELP coding at 4800 bps. raising the sampling rate by the same factor as the corresponding down-sampling box) followed by a bandpass filter similar to the analysis filter. ˆ Such perfect reconstruction filter bank systems are the basis for a wideranging class of speech and audio coders called subband coders [25]. but by careful ˆ . On the far right in the diagram is a set of “bandpass interpolators” where each interpolator is composed of an up-sampler (i.13 depicts the general nature of this large class of coding systems. we have really only focused on a few of the most important examples that have current relevance. Figure 7.132 Digital Speech Coding purpose is to reduce “peakiness” due to the minimum-phase nature of the linear predictive vocal tract system.e. and the outputs of the down-samplers are connected to the corresponding inputs of the bandpass interpolators. 1200. 7. On the far left of the diagram is a set (bank) of analysis bandpass filters. and 600 bps. but these add only slightly to either the analysis computation or the bit rate [80]. its superior performance led to a new Department of Defense standard in 1996 and subsequently to MIL-STD-3005 and NATO STANAG 4591. These are called frequency-domain coders because they are based on the principle of decomposing the speech signal into individual frequency bands. These modifications to the basic two-state excitation LPC vocoder produce marked improvements in the quality of reproduction of the speech signal. When the outputs of the filter bank are quantized by some sort of quantizer. If the filters are carefully designed.

of course. . To see how subband coders work. This masking was incorporated in some of the coding systems discussed so far. the filter bank structure incorporates important features of the speech perception model. the quantization error affects all frequencies in 23 The total bit rate will be the sum of the products of the sampling rates of the downsampled channel signals times the number of bits allocated to each channel. which incorporated the speech production model into the quantization process. but ADPCM could be used in principle. The goal with such coders is. adaptive PCM is used. As discussed in Chapter 3.7. Typically. design of the quantizer the output can be indistinguishable from the input perceptually. The quantizers can be any quantization operator that preserves the waveform of the signal.5 Frequency-Domain Coders 133 Fig. Because the down-sampled channel signals are full band at the lower sampling rate. 7. suppose first that the individual channels are quantized independently. In contrast to the coders that we have already discussed. the basilar membrane effectively performs a frequency analysis of the sound impinging on the eardrum. but only in a rudimentary manner. The coupling between points on the basilar membrane results in the masking effects that we mentioned in Chapter 3. for the total composite bit rate23 to be as low as possible while maintaining high quality.13 Subband coder and decoder for speech audio.

bands with low energy can be encoded with correspondingly low absolute error. The simplest approach is to pre-assign a fixed number of bits to each channel. Just as important. as suggested by the dotted boxes. Another approach is to allocate bits among the channels dynamically according to a perceptual criterion. a particular band can be quantized according to its perceptual importance. bits are allocated among the channels in an iterative process so that as much of the quantization error is inaudible as possible for the total bit budget.6 Evaluation of Coders In this chapter. and cepstrum analysis and also on models .13 if. This typically involves the computation of a fine-grained spectrum analysis using the DFT. vector quantization can be used effectively [23]. This is the basis for modern audio coding standards such as the various MPEG standards. [25] found that subband coders operating at 16 kbps and 9. if the channel bandwidths are all the same so that the output samples of the down-samplers can be treated as a vector. but the important point is that no other bands are affected by the errors in a given band. we have discussed a wide range of options for digital coding of speech signals. Such systems are based on the principles depicted in Figure 7. Much more sophisticated quantization schemes are possible in the configuration shown in Figure 7. Furthermore. What is not shown in that figure is additional analysis processing that is done to determine how to allocate the bits among the channels. respectively. 7. For example. From this spectrum it is possible to determine which frequencies will mask other frequencies. the quantization of the channels is done jointly. Using this audibility threshold. The details of perceptual audio coding are presented in the recent textbook by Spanias [120]. filter banks. and on the basis of this a threshold of audibility is determined.13. These systems rely heavily on the techniques of linear prediction. Crochiere et al. In an early non-uniformly spaced five channel system.6 kbps were preferred by listeners to ADPCM coders operating at 22 and 19 kbps.134 Digital Speech Coding that band.

Throughout this chapter we have mentioned bit rates. and quality as important practical dimensions of the speech coding problem. however. increasing the bit rate of an open-loop two-state-excitation LPC vocoder above about 2400 bps does not improve quality very much. robustness to transmission errors. improved pitch detection and source modeling can improve quality in an LPC vocoder as witnessed by the success of the MELP system. In the final analysis. Fortunately.6 Evaluation of Coders 135 for speech production and speech perception. the choice of a speech coder will depend on constraints imposed by the application such as cost of coder/decoder. it is important to note that for many systems. but lowering the bit rate causes noticeable degradation. there are many possibilities to choose from. available transmission capacity. . On the other hand. increasing the bit rate indefinitely does not necessarily continue to improve quality.7. Most coding schemes have many operations and parameters that can be chosen to tradeoff among the important factors. increasing the bit rate will improve quality of reproduction. complexity of computation. but this generally would come with an increase in computation and processing delay. All this knowledge and technique is brought to bear on the problem of reducing the bit rate of the speech representation while maintaining high quality reproduction of the speech signal. and quality requirements. In general. For example. processing delay.

The input to the TTS system is text and the output is synthetic speech. The two fundamental processes performed by all TTS systems are text analysis (to determine the abstract underlying linguistic description of the speech) and speech synthesis (to produce the speech sounds corresponding to the text input). Such systems are often referred to as text-to-speech synthesis (or TTS) systems. we will be concerned with computer simulation of the upper part of the speech chain in Figure 1.2. 8. In other words.8 Text-to-Speech Synthesis Methods In this chapter. 136 . and their general structure is illustrated in Figure 8.1. Fig.1 Block diagram of general TTS system. we discuss systems whose goal is to convert ordinary text messages into intelligible and natural sounding synthetic speech so as to transmit information from a machine to a human user.

The basic text processing benefits from an online dictionary of word pronunciations along with rules for determining word etymology.g.1 Text Analysis 137 8. depending on context).1 must determine three things from the input text string. and finally performing a linguistic analysis to determine grammatical information about words and phrases within the text. . Figure 8.1 Text Analysis The text analysis module of Figure 8. and how much emphasis should be given to individual words and phrases within the final spoken output speech (3) semantic focus and ambiguity resolution: the text analysis process must resolve homographs (words that are spelled alike but can be pronounced differently. the intonation of the speech.8. where the tags denote the linguistic properties of the words of the input text string.2 shows more detail on how text analysis is performed. what rate of speaking is most appropriate for the material being spoken. email message versus paragraph of text from an encyclopedia article). and the duration of each of the sounds in the utterance (2) syntactic structure of the sentence to be spoken: the text analysis process must determine where to place pauses. The input to the analysis is plain English text. The output of the basic text processing step is tagged text. normalizing the text (so as to determine how to pronounce words like proper names or homographs with multiple pronunciations).. namely: (1) pronunciation of the text string: the text analysis process must decide on the set of phonemes that is to be spoken. including detecting the structure of the document containing the text (e. The first stage of processing does some basic text processing operations. the degree of stress at various points in speaking. and also must use rules to determine word etymology to decide on how best to pronounce names and foreign words and phrases.

1 Document Structure Detection The document structure detection module seeks to determine the location of all punctuation marks in the text. 8. an end of sentence marker is usually a period.2 Text Normalization Text normalization methods handle a range of text problems that occur in real applications of TTS systems.2 Components of the text analysis process.5 in. 8. . ?. in St. 8. Louis” Example 2: “She worked for DEC in Maynard MA” . However this is not always the case as in the sentence.. including how to handle abbreviations and acronyms as in the following sentences: Example 1: “I live on Bourbon St. neither of which denote the end of the sentence.138 Text-to-Speech Synthesis Methods Fig. or an exclamation point. a question mark.1. For example. long” where there are two periods. !. “This car is 72. and to decide their significance with regard to the sentence and paragraph structure of the input text.1.

etc.e. the text “St. times currency.e. but typically a simple. depending on the context.2 is a linguistic analysis of the input text. A conventional parser could be used as the basis of the linguistic analysis of the printed text. Digital Equipment Corporation.” and dates.3 Linguistic Analysis The third step in the basic text processing block of Figure 8..8..1. account numbers. Thus the string “$10. i. especially those from languages other than English. for prominence in the sentence • the style of speaking. the use of a pronoun to refer back to another word unit) • the word (or words) on which emphasis are to be placed.50” should be pronounced as “ten dollars and fifty cents” rather than as a sequence of characters. relaxed. etc... One other important text normalization problem concerns the pronunciation of proper names. where a pause in speaking might be appropriate • the presence of anaphora (e. . for each word in the printed string.g. the following linguistic properties: • the part of speech (POS) of the word • the sense in which each word is used in the current context • the location of phrases (or phrase groups) with a sentence (or paragraph). nine hundred. i. and twenty.” is pronounced as street or saint. in Example 1. irate.1 Text Analysis 139 where.g. shallow analysis is performed since most linguistic parsers are very slow. but is virtually never pronounced as the letter sequence “D E C. e. with the goal of determining. 8. and in Example 2 the acronym DEC can be pronounced as either the word “deck” (the spoken acronym) or the name of the company. emotional.” Other examples of text normalization include number strings like “1920” which can be pronounced as the year “nineteen twenty” or the number “one thousand.

the conversion to sounds is straightforward and the dictionary search begins on the next word. Although there are a variety of ways of performing this analysis. The basis for this is the context in which the word occurs. namely conversion from the text to (marked) speech sounds. One simple example of homograph disambiguation is seen in the phrases “an absent boy” versus the sentence “do you choose to absent yourself. If not. the dictionary search attempts to find affixes (both prefixes and suffixes) and strips them . 8.1. perhaps the most straightforward method is to rely on a standard pronunciation dictionary. The phonetic analysis block of Figure 8. along with a set of letter-to-sound rules for words outside the dictionary. both locally (emphasis) and globally (speaking style).4 Phonetic Analysis Ultimately.5 Homograph Disambiguation The homograph disambiguation operation must resolve the correct pronunciation of each word in the input string that has more than one pronunciation.1.140 Text-to-Speech Synthesis Methods 8. If so. in the second phrase the word “absent” is a verb and the accent is on the second syllable.2 provides the processing that enables the TTS system to perform this conversion. 8. in its entirety. with the help of a pronunciation dictionary.” In the first phrase the word “absent” is an adjective and the accent is on the first syllable. as is the case most often. in the word dictionary. The way in which these steps are performed is as follows.1.6 Letter-to-Sound (LTS) Conversion The second step of phonetic analysis is the process of grapheme-tophoneme conversion. Each individual word in the text string is searched independently.3 shows the processing for a simple dictionary search for word pronunciation. First a “whole word” search is initiated to see if the printed word exists. Figure 8. the tagged text obtained from the basic text processing block of a TTS system has to be converted to a sequence of tagged phones which describe both the sounds to be produced as well as the manner of speaking.

from the word attempting to find the “root form” of the word. The determination of the sequence of speech sounds is mainly performed by the phonetic analysis step as outlined above.7 Prosodic Analysis The last step in the text analysis system of Figure 8. namely the sequence of speech sounds.1 Text Analysis 141 Fig. their durations.2 is prosodic analysis which provides the speech synthesizer with the complete set of synthesis controls. If the root form is not present in the dictionary. and then does another “whole word” search. 8. The assignment of duration and pitch contours is done by a set of pitch .1.8. and an associated pitch contour (variation of fundamental frequency with time). a set of letter-to-sound rules is used to determine the best pronunciation (usually based on etymology of the word) of the root form of the word. again followed by the reattachment of stripped out affixes (including the case of no stripped out affixes).3 Block diagram of dictionary search for proper word pronunciation. 8.

The second generation of speech synthesis methods (from 1977 to 1992) was based primarily on an LPC representation of sub-word units such as diphones (half phones). 8.142 Text-to-Speech Synthesis Methods and duration rules. By carefully modeling and representing diphone units via LPC parameters. . 8. Although the intelligibility improved dramatically Fig.2 Evolution of Speech Synthesis Systems A summary of the progress in speech synthesis. The synthesis suffered from poor intelligibility and poor naturalness. During the first generation (between 1962 and 1977) formant synthesis of phonemes using a terminal analog synthesizer was the dominant technology using rules which related the phonetic decomposition of the sentence to formant frequency contours.4 Time line of progress in speech synthesis and TTS systems. along with a set of rules for assigning stress and determining where appropriate pauses should be inserted so that the local and global speaking rates appear to be natural. over the period 1962– 1997. it was shown that good intelligibility synthetic speech could be reliably obtained from text input by concatenating the appropriate diphone units.4. This figure shows that there have been 3 generations of speech synthesis systems. is given in Figure 8.

cs. diphones. etc.2.2 Evolution of Speech Synthesis Systems 143 over first generation formant synthesis. Both tasks.html.1. .1 Early Speech Synthesis Approaches Once the abstract underlying linguistic description of the text input has been determined via the steps of Figure 8. in which the method of “unit selection synthesis” was introduced and perfected. or syllables (2) choice of synthesis parameters: including LPC features. The third generation of speech synthesis technology was the period from 1992 to the present. articulatory parameters. sinusoidal parameters. phones. namely: (1) choice of synthesis units: including whole words. (3) method of computation: including rule-based systems or systems which rely on the concatenation of stored speech units. 8. We begin our discussion of speech synthesis approaches with a review of early systems. and then discuss unit selection methods of speech synthesis later in this chapter. The resulting synthetic speech from this third generation technology had good intelligibility and naturalness that approached that of human-generated speech. are difficult to achieve and depend critically on three issues in the processing of the speech synthesizer “backend” of Figure 8. the remaining (major) task of TTS systems is to synthesize a speech waveform whose intelligibility is very high (to make the speech useful as a means of communication between a machine and a human). dyads. A detailed survey of progress in text-to-speech conversion up to 1987 is given in the review paper by Klatt [65]. the naturalness of the synthetic speech remained low due to the inability of single diphone units to represent all possible combinations of sound using that diphone unit.2. formants. waveform templates.8. The synthesis examples that accompanied that paper are available for listening at http://www.indiana. namely attaining high intelligibility along with reasonable naturalness.edu/rhythmsp/ASA/Contents. and whose naturalness is as close to real speech as possible. primarily by Sagisaka at ATR Labs in Kyoto [108].

2. the words can be stored as sampled waveforms and simply concatenated in the correct sequence.5 Wideband spectrograms of a sentence spoken as a sequence of isolated words (top panel) and as a continuous speech utterance (bottom panel). since it does not take into account the “co-articulation” effects of producing phonemes in continuous speech.5.5. even for this trivial Fig.144 Text-to-Speech Synthesis Methods 8. distinct pauses between words) as shown at the top of Figure 8. It can be seen that. For greatest simplicity.5. and as a continuous utterance. but unnatural sounding speech. as shown at the bottom of Figure 8. Words spoken in continuous speech sentences are generally much shorter in duration than when spoken in isolation (often up to 50% shorter) as illustrated in Figure 8. 8.2 Word Concatenation Synthesis Perhaps the simplest approach to creating a speech utterance corresponding to a given text string is to literally splice together prerecorded words corresponding to the desired utterance. This figure shows wideband spectrograms for the sentence “This shirt is red” spoken as a sequence of isolated words (with short. . This approach generally produces intelligible. and it does not provide either for the adjustment of phoneme durations or the imposition of a desired pitch variation across the utterance.

The rational for this is that the parametric representations. shorter units such as phonemes or diphones are more suitable for synthesis. in reality this type of synthesis is not a practical approach.1 There are many reasons for this. etc. the duration of the continuous sentence is on the order of half that of the isolated word version.2 Evolution of Speech Synthesis Systems 145 example. and further the formant tracks of the continuous sentence do not look like a set of uniformly compressed formant tracks from the individual words. but a major problem is that there are far too many words to store in a word catalog for word concatenation synthesis to be practical except in highly restricted situations. The word concatenation approach can be made more sophisticated by storing the vocabulary words in a parametric form (formants. LPC parameters. Efforts to overcome the limitations of word concatenation followed two separate paths. . This requires that the control parameters for all the words in the task vocabulary (as obtained from a training set of words) be stored as representations of the words. being more closely related to the model for speech production. A special set of word concatenation rules is then used to create the control signals for the synthesizer. the boundaries between words in the upper plot are sharply defined. For example. As we will see.7 million distinct surnames in the United States and each of them would have to be spoken and stored for a general word concatenation synthesis method. there are about 1. One approach was based on controlling the motions of a physical model of the speech articulators based on the sequence of phonemes from the text analysis. Furthermore. can be manipulated so as to blend the words together.8. equally important limitation is that word-length segments of speech are simply the wrong sub-units. This requires sophisticated control 1 It should be obvious that whole word concatenation synthesis (from stored waveforms) is also impractical for general purpose synthesis of speech. and impose a desired pitch variation. shorten them. while they are merged in the lower plot. A second.) such as employed in the speech coders discussed in Chapter 7 [102]. Although such a synthesis system would appear to be an attractive alternative for general purpose synthesis of speech.

g.. velum. control parameters (e. jaw. etc. 8. formants and pitch) can be derived by applying the acoustic theory of speech production and then used to control a synthesizer such those used for speech coding. pitch period. From the vocal tract shapes and sources produced by the articulatory model. Such models were thought to be inherently more natural than vocal tract analog models since: • we could impose known and fairly well understood physical constraints on articular movements to create realistic motions of the tongue.1 [22]. What we have learned about articulatory modeling of speech is that it requires a highly accurate model of the vocal cords and of the vocal tract for the resulting synthetic speech quality to be considered acceptable. teeth.). Again.3 Articulatory Methods of Synthesis Articulatory models of human speech are based upon the application of acoustic theory to physical models such as depicted in highly stylized form in Figure 2. etc. the rules (algorithm) for computing the control parameters are mostly derived by empirical means. formant parameters.146 Text-to-Speech Synthesis Methods rules that are mostly derived empirically.2.g. thereby potentially making the speech more natural sounding. LPC parameters. It further requires rules for handling the dynamics of the . either via direct methods (namely solving the wave equation). or indirectly by converting articulatory shapes to formants or LPC parameters • we could highly constrain the motions of the articulatory parameters so that only natural motions would occur. An alternative approach eschews the articulatory model and proceeds directly to the computation of the control signals for a source/system model (e.. • we could use X-ray data (MRI data today) to study the motion of the articulators in the production of individual speech sounds. thereby increasing our understanding of the dynamics of speech production • we could model smooth articulatory parameter motions between sounds.

2 Evolution of Speech Synthesis Systems 147 articulator motion in the context of the sounds being produced. and a transfer function of the human vocal tract in the form of a rational system function. . the terms analog and terminal result from the historical context of early speech synthesis studies. This synthesis process has been called terminal analog synthesis because it is based on a model (analog) of the human vocal tract production of speech that seeks to produce a signal at its output terminals that is equivalent to the signal produced by a human talker.) 2 In the designation “terminal analog synthesis”. −k (8.2 and 4. In this approach.4 Terminal Analog Synthesis of Speech The alternative to articulatory synthesis is called terminal analog speech synthesis. i. This can be confusing since today. . bq } and {ak } = {a1 .e. . each sound of the language (phoneme) is characterized by a source excitation function and an ideal vocal tract model. . ap } are the (timevarying) coefficients of the vocal tract filter.8. . E(z) is the z-transform of the vocal tract excitation signal. b1 . b2 . . and thus. . e[n]. a2 .. and {bk } = {b0 .2 The basis for terminal analog synthesis of speech is the source/system model of speech production that we have used many times in this text. . Speech is produced by varying (in time) the excitation and the vocal tract model control parameters at a rate commensurate with the sounds being produced. q b0 + H(z) = B(z) S(z) = = E(z) A(z) k=1 p bk z −k . So far we have been unable to learn all such rules. e[n].2. “analog” implies “not digital” as well as an analogous thing. “Terminal” originally implied the “output terminals” of an electronic analog (not digital) circuit or system that was an analog of the human speech production system. . s[n].1) 1− k=1 ak z where S(z) is the z-transform of the output speech signal. (See Figures 2. namely an excitation source. articulatory speech synthesis methods have not been found to be practical for synthesizing speech of acceptable quality.1. 8.

111]. implemented via a parallel branch is adequate.6 Speech synthesizer based on a cascade/serial (formant) synthesis model. a time-varying voiced signal gain. The voiced speech (upper) branch includes an impulse generator (controlled by a time-varying pitch period. F2 . it has been shown that it is preferable to factor the numerator and denominator polynomials into either a series of cascade (serial) resonances. F1 . . It has also been shown that an all-pole model (B(z) = constant) is most appropriate for voiced (non-nasal) speech.6 [95. P0 ). Finally a fixed spectral compensation network can be used to model the combined effects of glottal pulse shape and radiation characteristics from the lips and mouth. However. Fig. AV . F3 ). based on the above discussion. For unvoiced speech a simpler model (with one complex pole and one complex zero). is shown in Figure 8.1) can be implemented as a speech synthesis system using a direct form implementation. The vocal tract representation of (8. and one fixed resonance F4 . both the excitation signal properties (P0 and V /U V ) and the filter coefficients of (8.1) change periodically so as to synthesize different phonemes. A complete serial terminal analog speech synthesis model. and an all-pole discrete-time system that consists of a cascade of 3 time-varying resonances (the first three formants. based on two real poles in the z-plane. 8.148 Text-to-Speech Synthesis Methods For most practical speech synthesis systems. or into a parallel combination of resonances.

independent of the pitch period. and a resonance/antiresonance system consisting of a time-varying pole (FP ) and a time-varying zero (FZ ). using unit selection methods. it remains a very challenging task to compute the synthesis parameters. The voiced and unvoiced components are added and processed by the fixed spectral compensation network to provide the final synthetic speech output.6 is highly variable with explicit model shortcomings due to the following: • voiced fricatives are not handled properly since their mixed excitation is not part of the model of Figure 8.8. a time-varying unvoiced signal gain.3 Unit Selection Methods The key idea of a concatenative TTS system.6 are alleviated by a more complex model proposed by Klatt [64]. However. Nevertheless Klatt’s Klattalk system achieved adequate quality by 1983 to justify commercialization by the Digital Equipment Corporation as the DECtalk system. The resulting quality of the synthetic speech produced using a terminal analog synthesizer of the type shown in Figure 8. Many of the short comings of the model of Figure 8. although the unit selection methods to be discussed in Section 8. 8. is inadequate and produces buzzy sounding voiced speech • the spectral compensation model is inaccurate and does not work well for unvoiced sounds. Some DECtalk systems are still in operation today as legacy systems. AN . is to use synthesis segments that are sections of prerecorded .3 now provide superior quality in current applications. even with a more sophisticated synthesis model.3 Unit Selection Methods 149 The unvoiced speech (lower) branch includes a white noise generator.6 • nasal sounds are not handled properly since nasal zeros are not included in the model • stop consonants are not handled properly since there is no precise timing and control of the complex excitation signal • use of a fixed pitch pulse shape.

150 Text-to-Speech Synthesis Methods natural speech [31, 50, 108]. The word concatenation method discussed in Section 8.2.2 is perhaps the simplest embodiment of this idea; however, as we discussed, shorter segments are required to achieve better quality synthesis. The basic idea is that the more segments recorded, annotated, and saved in the database, the better the potential quality of the resulting speech synthesis. Ultimately, if an infinite number of segments were recorded and saved, the resulting synthetic speech would sound natural for virtually all possible synthesis tasks. Concatenative speech synthesis systems, based on unit selection methods, are what is conventionally known as “data driven” approaches since their performance tends to get better the more data that is used for training the system and selecting appropriate units. In order to design and build a unit selection system based on using recorded speech segments, several issues have to be resolved, including the following: (1) What speech units should be used as the basic synthesis building blocks? (2) How the synthesis units are selected (extracted) from natural speech utterances? (3) How the units are labeled for retrieval from a large database of units? (4) What signal representation should be used to represent the units for storage and reproduction purposes? (5) What signal processing methods can be used to spectrally smooth the units (at unit junctures) and for prosody modification (pitch, duration, amplitude)? We now attempt to answer each of these questions. 8.3.1 Choice of Concatenation Units

The units for unit selection can (in theory) be as large as words and as small as phoneme units. Words are prohibitive since there are essentially an infinite number of words in English. Subword units include syllables (about 10,000 in English), phonemes (about 45 in English, but they are highly context dependent), demi-syllables (about 2500

8.3 Unit Selection Methods

151

in English), and diphones (about 1500–2500 in English). The ideal synthesis unit is context independent and easily concatenates with other (appropriate) subword units [15]. Based on this criterion, the most reasonable choice for unit selection synthesis is the set of diphone units.3 Before any synthesis can be done, it is necessary to prepare an inventory of units (diphones). This requires significant effort if done manually. In any case, a large corpus of speech must be obtained from which to extract the diphone units as waveform snippets. These units are then represented in some compressed form for efficient storage. High quality coding such as MPLP or CELP is used to limit compression artifacts. At the final synthesis stage, the diphones are decoded into waveforms for final merging, duration adjustment and pitch modification, all of which take place on the time waveform. 8.3.2 From Text to Diphones

The synthesis procedure from subword units is straightforward, but far from trivial [14]. Following the process of text analysis into phonemes and prosody, the phoneme sequence is first converted to the appropriate sequence of units from the inventory. For example, the phrase “I want” would be converted to diphone units as follows: Text Input: I want. Phonemes: /#/ AY/ /W/ /AA/ /N/ /T/ /#/ Diphones: /# AY/ /AY-W/ /W-AA/ /AA-N/ /N-T/ /T- #/, where the symbol # is used to represent silence (at the beginning and end of each sentence or phrase). The second step in online synthesis is to select the most appropriate sequence of diphone units from the stored inventory. Since each diphone unit occurs many times in the stored inventory, the selection
3A

diphone is simply the concatenation of two phonemes that are allowed by the language constraints to be sequentially contiguous in natural speech. For example, a diphone based upon the phonemes AY and W would be denoted AY−W where the − denotes the joining of the two phonemes.

152 Text-to-Speech Synthesis Methods of the best sequence of diphone units involves solving a dynamic programming search for the sequence of units that minimizes a specified cost function. The cost function generally is based on diphone matches at each of the boundaries between diphones, where the diphone match is defined in terms of spectral matching characteristics, pitch matching characteristics, and possibly phase matching characteristics. 8.3.3 Unit Selection Synthesis

The “Unit Selection” problem is basically one of having a given set of target features corresponding to the spoken text, and then automatically finding the sequence of units in the database that most closely match these features. This problem is illustrated in Figure 8.7 which shows a target feature set corresponding to the sequence of sounds (phonemes for this trivial example) /HH/ /EH/ /L/ /OW/ from the word /hello/.4 As shown, each of the phonemes in this word have multiple representations (units) in the inventory of sounds, having been extracted from different phonemic environments. Hence there are many versions of /HH/ and many versions of /EH/ etc., as illustrated in Figure 8.7. The task of the unit selection module is to choose one of each of the multiple representations of the sounds, with the goal being to minimize the total perceptual distance between segments of the chosen sequence, based on spectral, pitch and phase differences throughout the sounds and especially at the boundary between sounds. By specifying costs associated with each unit, both globally across the unit,
“Trained” Perceptual Distance

HH

EH

L

OW

•••

HH HH HH HH

EH

EH E H

EH EH EH

L L L L L L

OW O W OW OW

Fig. 8.7 Illustration of basic process of unit selection. (After Dutoit [31].)
4 If

diphones were used, the sequence would be /HH-EH/ /EH-L/ /L-OW/.

throughout the unit. 99]. This optimal sequence (or equivalently the optimal path through the combination of all possible versions of each unit in the string) can be found using a Viterbi search (dynamic programming) [38. namely a nodal cost based on the unit segmental distortion (USD) which is defined as the difference between the desired spectral pattern of the target (suitably defined) and that of the candidate unit. and locally at the unit boundaries with adjacent units. By way of example. consider a target context of the word “cart” with target phonemes /K/ /AH/ /R/ /T/. This dynamic programming search process is illustrated in Figure 8. where the transitional costs (the arcs) reflect the cost of concatenation of a pair of units based on acoustic distances. and the nodes represent the target costs based on the linguistic identity of the unit. The Viterbi search effectively computes the cost of every possible path through the lattice and determines the path with the lowest total cost. 8.8 for a 3 unit search.8 Online unit selection based on a Viterbi search through a lattice of alternatives. Thus there are two costs associated with the Viterbi search.3 Unit Selection Methods 153 Fig.8. we can find the sequence of units that best “join each other” in the sense of minimizing the accumulated distance across the sequence of units. and the unit concatenative distortion (UCD) which is defined as the spectral (and/or pitch and/or phase) discontinuity across the boundaries of the concatenated units. and a source context of the phoneme /AH/ obtained from the .

154 Text-to-Speech Synthesis Methods

Fig. 8.9 Illustration of unit selection costs associated with a string of target units and a presumptive string of selected units.

source word “want” with source phonemes /W/ /AH/ /N/ /T/. The USD distance would be the cost (specified analytically) between the sound /AH/ in the context /W/ /AH/ /N/ versus the desired sound in the context /K/ /AH/ /R/. Figure 8.9 illustrates concatenative synthesis for a given string of target units (at the bottom of the figure) and a string of selected units from the unit inventory (shown at the top of the figure). We focus our attention on Target Unit tj and Selected Units θj and θj+1 . Associated with the match between units tj and θj is a USD Unit Cost and associated with the sequence of selected units, θj and θj+1 , is a UCD concatenation cost. The total cost of an arbitrary string of N selected units, Θ = {θ1 , θ2 , . . . , θN }, and the string of N target units, T = {t1 , t2 , . . . , tN } is defined as:
N N −1

d(Θ, T ) =
j=1

du (θj , tj ) +
j=1

dt (θj , θj+1 ),

(8.2)

where du (θj , tj ) is the USD cost associated with matching target unit tj with selected unit θj and dt (θj , θj+1 ) is the UCD cost associated with concatenating the units θj and θj+1 . It should be noted that when selected units θj and θj+1 come from the same source sentence and are adjacent units, the UCD cost goes to zero, as this is as natural a concatenation as can exist in the database. Further, it is noted that generally there is a small overlap region between concatenated units. The optimal path (corresponding to the optimal sequence of units) can be efficiently computed using a standard Viterbi search ([38, 99]) in

8.3 Unit Selection Methods

155

which the computational demand scales linearly with both the number of target and the number of concatenation units. The concatenation cost between two units is essentially the spectral (and/or phase and/or pitch) discontinuity across the boundary between the units, and is defined as:
p+2

dt (θj , θj+1 ) =
k=1

wk Ck (θj , θj+1 ),

(8.3)

where p is the size of the spectral feature vector (typically p = 12 mfcc coefficients (mel-frequency cepstral coefficients as explained in Section 5.6.3), often represented as a VQ codebook vector), and the extra two features are log power and pitch. The weights, wk are chosen during the unit selection inventory creation phase and are optimized using a trial-and-error procedure. The concatenation cost essentially measures a spectral plus log energy plus pitch difference between the two concatenated units at the boundary frames. Clearly the definition of concatenation cost could be extended to more than a single boundary frame. Also, as stated earlier, the concatenation cost is defined to be zero (dt (θj , θj+1 ) = 0) whenever units θj and θj+1 are consecutive (in the database) since, by definition, there is no discontinuity in either spectrum or pitch in this case. Although there are a variety of choices for measuring the spectral/log energy/pitch discontinuity at the boundary, a common cost function is the normalized mean-squared error in the feature parameters, namely: Ck (θj , θj+1 ) =
θ

[fk j (m) − fk j+1 (1)]2 , 2 σk

θ

θ

(8.4)

where fk j (l) is the kth feature parameter of the lth frame of segment θj , 2 m is the (normalized) duration of each segment, and σk is the variance of the kth feature vector component. The USD or target costs are conceptually more difficult to understand, and, in practice, more difficult to instantiate. The USD cost associated with units θj and tj is of the form:
q

du (θj , tj ) =
i=1

t wi φi Ti (fi j ), Ti (fi j ) ,

θ

t

(8.5)

156 Text-to-Speech Synthesis Methods where q is the number of features that specify the unit θj or tj , t wi , i = 1, 2, . . . , q is a trained set of target weights, and Ti (·) can be either a continuous function (for a set of features, fi , such as segmental pitch, power or duration), or a set of integers (in the case of categorical features, fi , such as unit identity, phonetic class, position in the syllable from which the unit was extracted, etc.). In the latter case, φi can be looked up in a table of distances. Otherwise, the local distance function can be expressed as a simple quadratic distance of the form φi Ti (fi j ), Ti (fi j ) = Ti (fi j ) − Ti (fi j )
θ t θ t 2

.

(8.6)

t The training of the weights, wi , is done off-line. For each phoneme in each phonetic class in the training speech database (which might be the entire recorded inventory), each exemplar of each unit is treated as a target and all others are treated as candidate units. Using this training set, a least-squares system of linear equations can be derived from which the weight vector can be solved. The details of the weight training methods are described by Schroeter [114]. The final step in the unit selection synthesis, having chosen the optimal sequence of units to match the target sentence, is to smooth/modify the selected units at each of the boundary frames to better match the spectra, pitch and phase at each unit boundary. Various smoothing methods based on the concepts of time domain harmonic scaling (TDHS) [76] and pitch synchronous overlap add (PSOLA) [20, 31, 82] have been proposed and optimized for such smoothing/modification at the boundaries between adjacent diphone units.

8.4

TTS Applications

Speech technology serves as a way of intelligently and efficiently enabling humans to interact with machines with the goal of bringing down cost of service for an existing service capability, or providing new products and services that would be prohibitively expensive without the automation provided by a viable speech processing interactive

The biggest problem with most TTS systems is that they have no idea as to how things should be said.. e. reading e-mail and fax messages) • voice portal providing voice access to web-based services • e-commerce agents • customized News. .. the more natural sounding these systems will become. The more TTS systems learn how to produce context-sensitive pronunciations of words (and phrases).5 TTS Future Needs Modern TTS systems have the capability of producing highly intelligible. and surprisingly natural speech utterances. so long as the utterance is not too long. or too technical. Hence. gas stations. driving directions. Sports scores. Stock Reports.g.8. cellphone) • a means of replacing expensive recorded Interactive Voice Response system prompts (a service which is particularly valuable when the prompts change often during the course of the day. alerting you to locations of restaurants. by way of example. prosody. but instead rely on the text analysis for emphasis. e.g.. or too syntactically complicated..5 TTS Future Needs 157 system. • giving voice to small and embedded devices for reporting information and alerts. traffic reports) • unified messaging (e.g. Examples of existing services where TTS enables significant cost reduction are as follows: • a dialog component for customer care applications • a means for delivering text messages over an audio connection (e. etc. Similarly some examples of new products and services which are enabled by a viable TTS technology are the following: • location-based services. stores in the vicinity of your current location • providing information in cars (e. 8..g. stock price quotations). phrasing and all so-called suprasegmental features of the spoken utterance.g.

Also needed in the unit selection process is a better spectral distance measure that incorporates human perception measures so as to find the best sequence of units for a given utterance. not to Mary or Bob. not the photos or the apple.e.. spectral properties) and actual features of a candidate recorded unit. . Finally.e. i. I gave the book to John. duration.e. I gave the book to John.e. The second future need of TTS systems is improvements in the unit selection process so as to better capture the target cost for mismatch between predicted unit specification (i. pitch. I gave the book to John.. i. better signal processing would enable improved compression of the units database.. thereby making the footprint of TTS systems small enough to be usable in handheld and mobile devices.e. not someone else.. each with different emphasis on words in the utterance. phoneme name. i. i. I did it..158 Text-to-Speech Synthesis Methods the utterance “I gave the book to John” has at least three different semantic interpretations.

9. The driving factor behind research in machine recognition of speech has been the potentially huge payoff of providing services where humans interact solely with machines. That is. noisy room. Interestingly. quiet office. the ultimate goal.e. which is in essence the inverse of the text-to-speech problem. the transducer or microphone). 159 . outdoors).. is to perform as well as a human listener. independent of the device used to record the speech (i. the speaker’s accent. or the acoustic environment in which the speaker is located (e. which has not yet been achieved.g..9 Automatic Speech Recognition (ASR) In this chapter.1 The Problem of Automatic Speech Recognition The goal of an ASR system is to accurately and efficiently convert a speech signal into a text message transcription of the spoken words. as a side benefit. thereby eliminating the cost of live agents and significantly reducing the cost of providing services. this process often provides users with a natural and convenient way of accessing information and services. we examine the process of speech recognition by machine.

In particular. which efficiently characterize the speech sounds. resulting in the speech waveform s[n]. W . s[n]. in the form of a sequence of words (possibly with pauses and other acoustic events such as uh’s. the speaker must compose a linguistically meaningful sentence. er’s etc. . X. This is in essence a digital simulation of the lower part of the speech chain diagram in Figure 1.2.160 Automatic Speech Recognition (ASR) Fig. the mel frequency cepstrum coefficients . X = {x1 .1. 9. Once the words are chosen. um’s. The feature vectors are computed on a frame-by-frame basis using the techniques discussed in the earlier chapters. To express that thought.1 and consist of an acoustic processor which analyzes the speech signal and converts it into a set of acoustic (spectral. xT }.2 shows a more detailed block diagram of the overall speech recognition system. which is a simplified version of the speech chain shown in Figure 1. It is assumed that the speaker intends to express some thought as part of a process of conversing with another human or with a machine. A simple conceptual model of the speech generation and speech recognition processes is given in Figure 9. We refer to the process of creating the speech waveform from the speaker’s intention as the Speaker Model since it reflects the speaker’s accent and choice of words to express a given thought or request. is converted to the sequence of feature vectors.2. the speaker sends appropriate control signals to the articulatory speech organs which form a speech utterance whose sounds are those required to speak the desired sentence. x2 . Figure 9. by the feature analysis block (also denoted spectral analysis). The processing steps of the Speech Recognizer are shown at the right side of Figure 9. .). temporal) features.1 Conceptual model of speech production and speech recognition processes. . . followed by a linguistic decoding process which makes a best (maximum likelihood) estimate of the words of the spoken sentence. resulting in the recognized sentence ˆ W . The input speech signal.

9. The remainder of this chapter is an attempt to give the flavor of what is involved in each part of Figure 9. an N -gram language model is used to compute a language model score for each proposed word string. Also. W that could have produced the input sequence of feature vectors.2 involves many details and.2 Building a Speech Recognition System The steps in building and evaluating a speech recognition system are the following: (1) choose the feature set and the associated signal processing for representing the properties of the speech signal over time . 9. The final block in the process is a confidence scoring process (also denoted as an utterance verification block). The pattern classification block (also denoted as the decoding and search block) decodes the sequence of feature vectors into a symbolic repreˆ sentation that is the maximum likelihood string. are widely used to represent the short-time spectral characteristics.2 Block diagram of an overall speech recognition system. which is used to provide a confidence score for each individual word in the recognized string. in some cases.2 Building a Speech Recognition System 161 Fig.2. The pattern recognition system uses a set of acoustic models (represented as hidden Markov models) and a word lexicon to provide the acoustic match score for each proposed string. 9. extensive digital computation. Each of the operations in Figure 9.

The pre-emphasized signal is next blocked into frames of N samples. Some type of cepstral bias removal is often used prior to calculation of the first and second order cepstral derivatives. Some of the important issues are summarized in this section. with frame shifts of 10 ms being most common.162 Automatic Speech Recognition (ASR) (2) choose the recognition task. with adjacent frames spaced M samples apart. Following (optionally used) simple noise removal methods. the basic speech sounds to represent the vocabulary (the speech units). hence adjacent frames overlap by 5–30 ms depending on the chosen values of N and M . articulatory.1 Recognition Feature Set There is no “standard” set of features for speech recognition. the spectral coefficients are normalized and converted to mel-frequency cepstral coefficients via standard analysis methods of the type discussed in Chapter 5 [27]. The analog speech signal is sampled and quantized at rates between 8000 up to 20. and the task semantics (if any) (3) train the set of speech acoustic and language models (4) evaluate performance of the resulting speech recognition system. Each of these steps may involve many choices and can involve significant research and development effort. 9.3.000 samples/s. Instead. including the recognition word vocabulary (the lexicon). the task syntax or language model. Typical values for N and M correspond to frames of duration 15–40 ms.2. A block diagram of the signal processing used in most modern large vocabulary speech recognition systems is shown in Figure 9. various combinations of acoustic. . A Hamming window is applied to each frame prior to spectral analysis using either standard spectral analysis or LPC methods. A first order (highpass) pre-emphasis network (1 − αz −1 ) is used to compensate for the speech spectral falloff at higher frequencies and approximates the inverse to the mouth transmission frequency response. and auditory features have been utilized in a range of speech recognition systems. The most popular acoustic features have been the (LPC-derived) mel-frequency cepstrum coefficients and their derivatives.

9.2 Building a Speech Recognition System

163

Fig. 9.3 Block diagram of feature extraction process for feature vector consisting of mfcc coefficients and their first and second derivatives.

Typically the resulting feature vector is the set of cepstral coefficients, and their first- and second-order derivatives. It is typical to use about 13 mfcc coefficients, 13 first-order cepstral derivative coefficients, and 13 second-order derivative cepstral coefficients, making a D = 39 size feature vector. 9.2.2 The Recognition Task

Recognition tasks vary from simple word and phrase recognition systems to large vocabulary conversational interfaces to machines. For example, using a digits vocabulary, the task could be recognition of a string of digits that forms a telephone number, or identification code, or a highly constrained sequence of digits that form a password. 9.2.3 Recognition Training

There are two aspects to training models for speech recognition, namely acoustic model training and language model training. Acoustic model

164 Automatic Speech Recognition (ASR) training requires recording each of the model units (whole words, phonemes) in as many contexts as possible so that the statistical learning method can create accurate distributions for each of the model states. Acoustic training relies on accurately labeled sequences of speech utterances which are segmented according to the transcription, so that training of the acoustic models first involves segmenting the spoken strings into recognition model units (via either a Baum–Welch [10, 11] or Viterbi alignment method), and then using the segmented utterances to simultaneously build model distributions for each state of the vocabulary unit models. The resulting statistical models form the basis for the pattern recognition operations at the heart of the ASR system. As discussed in Section 9.3, concepts such as Viterbi search are employed in the pattern recognition process as well as in training. Language model training requires a sequence of text strings that reflect the syntax of spoken utterances for the task at hand. Generally such text training sets are created automatically (based on a model of grammar for the recognition task) or by using existing text sources, such as magazine and newspaper articles, or closed caption transcripts of television news broadcasts, etc. Other times, training sets for language models can be created from databases, e.g., valid strings of telephone numbers can be created from existing telephone directories. 9.2.4 Testing and Performance Evaluation

In order to improve the performance of any speech recognition system, there must be a reliable and statistically significant way of evaluating recognition system performance based on an independent test set of labeled utterances. Typically we measure word error rate and sentence (or task) error rate as a measure of recognizer performance. A brief summary of performance evaluations across a range of ASR applications is given in Section 9.4.

9.3

The Decision Processes in ASR

The heart of any automatic speech recognition system is the pattern classification and decision operations. In this section, we shall give a brief introduction to these important topics.

9.3 The Decision Processes in ASR

165

9.3.1

Mathematical Formulation of the ASR Problem

The problem of automatic speech recognition is represented as a statistical decision problem. Specifically it is formulated as a Bayes maximum a posteriori probability (MAP) decision process where we seek to find ˆ the word string W (in the task language) that maximizes the a posteriori probability P (W |X) of that string, given the measured feature vector, X, i.e., ˆ W = arg max P (W |X).
W

(9.1)

Using Bayes’ rule we can rewrite (9.1) in the form: P (X|W )P (W ) ˆ . W = arg max W P (X) (9.2)

Equation (9.2) shows that the calculation of the a posteriori probability is decomposed into two terms, one that defines the a priori probability of the word sequence, W , namely P (W ), and the other that defines the likelihood that the word string, W , produced the feature vector, X, namely P (X|W ). For all future calculations we disregard the denominator term, P (X), since it is independent of the word sequence W which is being optimized. The term P (X|W ) is known as the “acoustic model” and is generally denoted as PA (X|W ) to emphasize the acoustic nature of this term. The term P (W ) is known as the “language model” and is generally denoted as PL (W ) to emphasize the linguistic nature of this term. The probabilities associated with PA (X|W ) and PL (W ) are estimated or learned from a set of training data that have been labeled by a knowledge source, usually a human expert, where the training set is as large as reasonably possible. The recognition decoding process of (9.2) is often written in the form of a 3-step process, i.e., ˆ W = arg max PA (X|W ) PL (W ),
W

(9.3)

Step 3

Step 1

Step 2

where Step 1 is the computation of the probability associated with the acoustic model of the speech sounds in the sentence W , Step 2 is the computation of the probability associated with the linguistic model of the words in the utterance, and Step 3 is the computation associated

t = 1. 97.3. . are small (close to 0). as: W = w1 . T times the frame shift in ms) and each frame.4 shows a simple Q = 5-state HMM for modeling a whole word.2 The Hidden Markov Model The most widely used method of building acoustic models (for both phonemes and words) is the use of a statistical characterization known as Hidden Markov Models (HMMs) [33.4) where the speech signal duration is T frames (i. the HMM is also characterized by an explicit set of state transitions. of the form: X = {x1 . . . 9. 2. a45 . W .3) we need to more explicit about the relationship between the feature vector. Figure 9. Similarly we can express the optimally decoded word sequence. .166 Automatic Speech Recognition (ASR) with the search through all valid sentences in the task language for the maximum likelihood sentence. x2 . Each HMM state is characterized by a mixture density Gaussian distribution that characterizes the statistical behavior of the feature vectors within the states of the model [61. in the model. wM .0). In addition to the statistical feature densities within states.6) where there are assumed to be exactly M words in the decoded string. . (9. which specify the probability of making a transition from state i to state j at each frame. T is an acoustic feature vector of the form: xt = (xt1 . . a23 . and the word sequence W . . . . . . aij .5) that characterizes the spectral/temporal properties of the speech signal at time t and D is the number of acoustic features in each frame.. and the jump transitions. the feature vector. As discussed above.e. aii are large (close to 1. a12 . thereby defining the time sequence of the feature vectors over the duration of the word. . . xtD ) (9. xT }. 98]. w2 . 62]. (9. X. . . a34 . xt2 . X. xt . Usually the self-transitions. 69. is a sequence of acoustic observations corresponding to each of T frames of the speech. . In order to be more explicit about the signal processing and computations associated with each of the three steps of (9. .

The Baum–Welch method is a hill-climbing algorithm and is iterated until a stable alignment of models and speech is obtained. The complete HMM characterization of a Q-state word model (or a sub-word unit like a phoneme model) is generally written as λ(A. .9. covariances and mixture gains for the distributions in each model state. The initial model can be randomly chosen or selected based on a priori knowledge of the model 1 The Baum–Welch algorithm is also widely referred to as the forward–backward method. 99]. The details of the Baum–Welch procedure are beyond the scope of this chapter but can be found in several references on Speech Recognition methods [49. left-to-right. 11].4 Word-based. state observation probability density.1 This algorithm aligns each of the various words (or subword units) with the spoken inputs and then estimates the appropriate means. In order to train the HMM (i.5.. π) with state transition matrix A = {aij . j ≤ Q}. 1 ≤ i ≤ Q} with π1 set to 1 for the “left-to-right” models of the type shown in Figure 9. 1 ≤ j ≤ Q}. B = {bj (xt ). 1 ≤ i. HMM with 5 states.e. B. 9. a labeled training set of sentences (transcribed into words and sub-word units) is used to guide an efficient training procedure known as the Baum–Welch algorithm [10. and initial state distribution.3 The Decision Processes in ASR 167 Fig.4. The heart of the training procedure for re-estimating HMM model parameters using the Baum– Welch procedure is shown in Figure 9. An initial HMM model is used to begin the training process. learn the optimal model parameters) for each word (or sub-word) unit. π = {πi .

4. as illustrated in Fig. It is a simple matter to go from the HMM for a whole word. a middle state representing the heart of the sound.6. This simple 3-state HMM is a basic sub-word unit model with an initial state representing the statistical characteristics at the beginning of a sound. parameters.168 Automatic Speech Recognition (ASR) Fig. This process is iterated until no further improvement in probabilities occurs with each new iteration. A word model is made by concatenating the appropriate sub-word HMMs. to an HMM for a sub-word unit (such as a phoneme) as shown in Figure 9. .5 The Baum–Welch training procedure based on a given training set of utterances.6 Sub-word-based HMM with 3 states. The iteration loop is a simple updating procedure for computing the forward and backward model probabilities based on an input speech database (the training set of utterances) and then optimizing the model parameters to give an updated HMM. 9. 9. as shown in Figure 9. and an ending state representing the spectral characteristics at the end of the sound.

T . ww . wM with a sequence of feature vectors. or a Viterbi alignment procedure [38. aligns with the last state in the M th word model. The resulting alignment procedure is illustrated in Figure 9. and the last feature vector. X = {x1 . We are now ready to define the procedure for aligning a sequence of M word models. . We see the sequence of feature vectors along the horizontal axis and the concatenated sequence of word states along the vertical axis.9.7.8. xT . w1 . . exceeds the total number of model states. giving the word model for the word “is” (pronounced as /IH Z/). 132] for which . . An optimal alignment procedure determines the exact best matching sequence between word model states and feature vectors such that the first feature vector. but clearly the alignment procedure works for any size model for any word.3 The Decision Processes in ASR 169 Fig. so that every state has at least a single feature vector associated with that state. . . In general the composition of a word (from sub-word units) is specified in a word lexicon or dictionary. . . xT }.8. however once the word model has been built it can be used much the same as whole word models for training and for evaluating word strings for maximizing the likelihood as part of the speech recognition process. (For simplicity we show each word model as a 5-state HMM in Figure 9. Figure 9. x1 .7 Word-based HMM for the word /is/ created by concatenating 3-state subword models for the sub-word units /ih/ and /z/. which concatenates the 3-state HMM for the sound /IH/ with the 3-state HMM for the sound /Z/.) The procedure for obtaining the best alignment between feature vectors and model states is based on either using the Baum–Welch statistical alignment procedure (in which we evaluate the probability of every alignment path and add them up to determine the probability of the word string). subject to the constraint that the total number of feature vectors. aligns with the first state in the first word model. . 9. x2 .

8 is based on the ease of evaluating the probability of any alignment path using the Baum–Welch or Viterbi procedures. . x2 . . 9. . . 9. The utility of the alignment procedure of Figure 9. . given the observed acoustic vectors. we determine the single best alignment path and use the probability score along that path as the probability measure for the current word string.3. xT } came from the word sequence W = w1 . w2 . we need to compute the probability that the acoustic vector sequence X = {x1 . i. .e.3).170 Automatic Speech Recognition (ASR) Fig. . We now return to the mathematical formulation of the ASR problem and examine in more detail the three steps in the decoding Equation (9. . . wM (assuming each word is represented as an HMM) and perform this computation for all possible word sequences.8 Alignment of concatenated HMM word models with acoustic feature vectors based on either a Baum–Welch or Viterbi alignment procedure.3 Step 1 — Acoustic Modeling The function of the acoustic modeling step (Step 1) is to assign probabilities to the acoustic realizations of a sequence of words..

j. Within each state of each word there is a probability density that characterizes the statistical properties of the feature vectors in that state. The optimal segmentation of these feature vectors (frames) into the M words is shown by the sequence of boxes.7) as the product T PA (X|W ) = t=1 PA xt |wj(t) . . is learned during a training phase of the recognizer. wM ). xT }|w1 . The process of assigning individual speech frames to the appropriate word model in an utterance is based on an optimal alignment process between the concatenated sequence of word models and the sequence of feature vectors of the spoken input utterance being recognized. in the word sequence. . . . we can express (9. This alignment process is illustrated in Figure 9. x2 . each of which corresponds to one of the words in the utterance and its set of optimally matching feature vectors.8) where we associate each frame of X with a unique word and state. w2 .7) If we make the assumption that each frame. We have seen in the previous section that the probability density of each state. We assume that each word model is further decomposed into a set of states which reflect the changing statistical properties of the feature vectors over time for the duration of the word. i(t) (9. . which we denote as bj (xt ). . . we calculate the local probability PA xt |wj(t) given that we know the word from which frame t came. of the word model. 2. . . i(t) wj(t) . and we denote the states as Sj .3 The Decision Processes in ASR 171 This calculation can be expressed as: PA (X|W ) = PA ({x1 .9. is aligned with i(t) word i(t) and HMM model state j(t) via the function wj(t) and if we assume that each frame is independent of every other frame. Further.9 which shows the set of T feature vectors (frames) along the horizontal axis. i(t) . . and the set of M words (and word model states) along the vertical axis. N . and for each word. j = 1. . Using a mixture of Gaussian densities to characterize the statistical distribution of the feature vectors in each state. . xt . We assume that each word is represented by an N -state HMM model. (9.

We now return to the issue of the calculation of the probability of frame xt being associated with the j(t)th state of the i(t)th word in .172 Automatic Speech Recognition (ASR) Fig. for mixture k in state j. for mixture k for state j. with the constraint cjk ≥ 0. 9. µjk . Ujk . we get a state-based probability density of the form: K bj (xt ) = k =1 cjk N[xt . (9. and covariance matrix. µjk . (9. The density constraints are: K cjk = 1. and N is a Gaussian density function with mean vector.9 Illustration of time alignment process between unknown utterance feature vectors and set of M concatenated word models. ∞ −∞ k=1 1≤j≤N 1 ≤ j ≤ N.10) (9.9) where K is the number of mixture components in the density function. cjk is the weight of the kth mixture component in state j.11) bj (xt )d xt = 1 . Ujk ].

. The key point is that we assign probabilities to acoustic realizations of a sequence of words by using hidden Markov models of the acoustic feature vectors within words. which is calculated as PA xt |wj(t) = bj(t) (xt ). Even for modest size vocabularies of about 1000 words. State-of-theart systems use context-dependent phone models as the basic units of recognition [99]. j. the amount of training data required for word models is excessive.9) are.12) The computation of (9. The alternative to word models is to build acoustic-phonetic models for the 40 or so phonemes in the English language and construct the model for a word by concatenating (stringing together sequentially) the models for the constituent phones in the word (as represented in a word dictionary or lexicon).9. The parameters. we “train the system” and learn the parameters of the best acoustic models for each word (or more specifically for each sound that comprises each word). the mean vectors. Although we have been discussing acoustic models for whole words. We come back to these issues later in this section.12) is incomplete since we have ignored the computation of the probability associated with the links between word states. Using an independent (and orthographically labeled) set of training data. according to the mixture model of (9. The use of such sub-word acoustic-phonetic models poses no real difficulties in either training or when used to build up word models and hence is the most widely used representation for building word models in a speech recognition system. for each state of the model. the mixture weights. PA (xt |wj(t) ). it should be clear that for any reasonable size speech recognition task.3 The Decision Processes in ASR 173 the utterance. in the alignment between a given word and a set of feature vectors corresponding to that word. and the covariance matrix. it is impractical to create a separate acoustic model for every possible word in the vocabulary since each word would have to be spoken in every possible context in order to build a statistically reliable model of the density functions of (9. and we have also not specified how to determine the within-word state.9). i(t) i(t) i(t) (9.

w2 . Perhaps the most popular way of constructing the language model is through the use of a statistical N -gram word grammar that is estimated from a large training set of text utterances. . 60. is to enable the computation of the a priori probability. all valid text strings in the language and assigning appropriate probability scores to each string.13) (9. Thus we assume we can write the probability of the sentence W . or grammar. W . . If we make the assumption that the probability of a word in a sentence is conditioned on only the previous N − 1 words. consistent with the recognition task [59. . .174 Automatic Speech Recognition (ASR) 9.14) = n=1 PL (wn |wn−1 . . 106]. either from the task at hand or from a generic database with applicability to a wide range of tasks. There are many ways of building Language Models for specific tasks.) For every sentence in the training set. . The purpose of the language model. based on the likelihood of that sequence of words occurring in the context of the task being performed by the speech recognition system. of a word string. wn−N +1 ).3. according to an N -gram language model. we have the basis for an N -gram language model. PL (W ). Hence the probability of the text string W =“Call home” for a telephone number identification task is zero since that string makes no sense for the specified task. including: (1) statistical training from text databases transcribed from task-specific dialogs (a learning procedure) (2) rule-based learning of the formal grammar associated with the task (3) enumerating. as PL (W ) = PL (w1 . Assume we have a large text training set of word-labeled utterances. We now describe the way in which such a language model is built. (Such databases could include millions or even tens of millions of text sentences.4 Step 2 — The Language Model The language model assigns probabilities to sequences of words. by hand. we have a text file that identifies the words in that sentence. . wM ) M (9. . wn−2 . .

the probability that a word wn was preceded by the pair of words (wn−1 . wn−2 ) = C(wn−2 . the trigram of words) consisting of (wn−2 . we compute this quantity as P (wn |wn−1 . bigram of words) (wn−2 . wn−1 . wn−1 ) (9. thereby taking inordinate amounts of computing power to solve by heuristic methods. Finite State Network (FSN) methods have evolved that reduce the computational burden by orders of magnitude.10 which shows a word pronunciation network for the word /data/. . even for very large speech recognition problems [81]. to estimate word “trigram” probabilities (i. thereby enabling exact maximum likelihood solutions in computationally feasible times.e. and the weight is an estimate of the probability that the arc is utilized in the pronunciation of the word in context. wn−2 )).. wn ) . The basic concept of a finite state network transducer is illustrated in Figure 9.3 The Decision Processes in ASR 175 where the probability of a word occurring in the sentence only depends on the previous N − 1 words and we estimate this probability by counting the relative frequencies of N -tuples of words in the training set. 9. to find the one with the maximum likelihood of having been spoken.. C(wn−2 . Fortunately.e. Thus. wn−1 ) as it occurs in the text training set.3. wn ) as it occurs in the text training set. for example. wn−1 . Each arc in the state diagram corresponds to a phoneme in the word pronunciation network. wn ) is the frequency count of the word triplet (i. through the use of methods from the field of Finite State Automata Theory.15) where C(wn−2 .9. wn−1 ) is the frequency count of the word doublet (i.. We see that for the word /data/ there are four total pronunciations.e. wn−1 . The key problem is that the potential size of the search space can be astronomically large (for large vocabularies and high average word branching factor language models). and C(wn−2 .5 Step 3 — The Search Problem The third step in the Bayesian approach to automatic speech recognition is to search the space of all valid word sequences from the language model.

12. thereby minimize the amount of duplicate computation). (After Mohri [81]. Further. we can carry the process down to the level of HMM phones and HMM states.32 /T/ /AX/ — probability of 0. into a much smaller network via the method of weighted finite state transducers (WFST). minimization. model phones. and weight pushing) and network optimization. equivalently. We can continue the process of creating efficient FSNs for each word in the task vocabulary (the speech dictionary or lexicon).11 [81]. and even model phrases. 9.10 Word pronunciation transducer for four pronunciations of the word /data/. and then combine word FSNs into sentence FSNs using the appropriate language model. The combined FSN of the 4 pronunciations is a lot more efficient than using 4 separate enumerations of the word since all the arcs are shared among the 4 pronunciations and the total computation for the full FSN for the word /data/ is close to 1/4 the computation of the 4 variants of the same word. A simple example of such a WFST network optimization is given in Figure 9.176 Automatic Speech Recognition (ASR) Fig. which combine the various representations of speech and language and optimize the resulting network to minimize the number of search states (and.) namely (along with their (estimated) pronunciation probabilities): (1) (2) (3) (4) /D/ /D/ /D/ /D/ /EY/ /EY/ /AE/ /AE/ /D/ /AX/ — probability of 0. Ultimately we can compile a very large network of model states. the WFST uses a unified mathematical .08 /D/ /AX/ — probability of 0. making the process even more efficient.48 /T/ /AX/ — probability of 0. model words. Using the techniques of network combination (which include network composition. determinization.

1 represents a wide range of possibilities for the implementation of automatic speech recognition. testing and evaluation is a major part of speech recognition research and development. model words. (After Mohri [81]. and as we mentioned before.3 we describe the most widely used statistical pattern recognition techniques that are employed for mapping the sequence of feature vectors into a sequence of symbols or words. we suggest how the techniques of digital speech analysis discussed in Chapters 4–6 can be applied to extract a sequence of feature vectors from the speech signal. model phones. Many variations on this general theme have been investigated over the past 30 years or more. Using these methods. 9.4 Representative Recognition Performance 177 Fig.11 Use of WFSTs to compile a set of FSNs into a single optimized network to minimize redundancy in the network. and in Section 9.2. 9. and model phrases) was able to be compiled down to a mathematically equivalent model with 108 states that was readily searched for the optimum word string with no loss of performance or word accuracy. Table 9.1 gives a summary of a range of the performance of some speech recognition and natural language understanding systems .4 Representative Recognition Performance The block diagram of Figure 9. In Section 9. an unoptimized network with 1022 states (the result of the cross product of model states.1.9.) framework to efficiently compile a large network into a minimal representation that is readily searched using standard Viterbi decoding methods [38].

1 Word error rates for a range of speech recognition systems.000 210. Corpus Connected digit strings (TI Database) Connected digit strings (AT&T Mall Recordings) Connected digit strings (AT&T HMIHY c ) Resource management (RM) Airline travel information system (ATIS) North American business (NAB & WSJ) Broadcast News Switchboard Call-home Type of speech Spontaneous Spontaneous Vocabulary size 11 (0–9.178 Automatic Speech Recognition (ASR) Table 9.3% for a very clean recording environment for the TI (Texas Instruments) connected digits database [68]).1 also shows the word error rates for a range of DARPA tasks ranging from • read speech of commands and informational requests about a naval ships database (the resource management system or RM) with a 1000 word vocabulary and a word error rate of 2.5 6.0 Conversational Read speech Spontaneous Read text Narrated News Telephone conversation Telephone conversation 11 (0–9.000 5.0 2. oh) 11 (0–9.0% and when embedded within conversational speech (the AT&T HMIHY (How May I Help You) c system) the word error rate increases significantly to 5.5% . the word error rates are very low (0.0%. oh) Word error rate (%) 0. Table 9. and application contexts [92].3 2. oh) 1000 2500 64. or ATIS [133]) with a 2500 word vocabulary and a word error rate of 2.000 28. speaking styles.6 ≈15 ≈27 ≈35 that have been developed so far. showing the lack of robustness of the recognition system to noise and other background disturbances [44].0% • spontaneous speech input for booking airlines travel (the airline travel information system.000 45. It can be seen that for a vocabulary of 11 digits. This table covers a range of vocabulary sizes. but when the digit strings are spoken in a noisy shopping mall environment the word error rate rises to 2.0 2.

Before ASR systems become ubiquitous in society. ASR systems fall far short of human speech perception in all but the simplest. or NAB) with a vocabulary of 64.000 word vocabulary and a word error rate of about 15% • recorded live telephone conversations between two unrelated individuals (the switchboard task [42]) with a vocabulary of 45. . most constrained tasks.6% • narrated news broadcasts from a range of TV news providers like CNBC. and even sentences so as to maintain an intelligent dialog with a customer. and robustness in order to utilize the technology for a wide range of tasks.9. many improvements will be required in both system performance and operational performance.000 words and a word error rate of 6. In the operational area we need better methods of detecting when a person is speaking to a machine and isolating the spoken input from the background.5 Challenges in ASR Technology 179 • read text from a range of business magazines and newspapers (the North American business task.5 Challenges in ASR Technology So far. and under a wide range of operating conditions. In the system area we need large improvements in accuracy. we need to be able to handle users talking over the voice prompts (so-called barge-in conditions). phrases. on a wide range of processors. and finally we need better methods of confidence scoring of words. we need more reliable and accurate utterance rejection methods so we can be sure that a word needs to be repeated when poorly recognized the first time. (the broadcast news task) with a 210.000 words and a word error rate of about 35%. efficiency. 9.000 words and a word error rate of about 27%. and a separate task for live telephone conversations between two family members (the call home task) with a vocabulary of 28.

A firm understanding of the fundamentals of acoustics. signal processing. and our understanding of the processing of speech in the human brain is at an even lower level. synthesized speech response systems. Digital speech processing systems have permeated society in the form of cellular speech coders. and speech recognition and understanding systems that handle a wide range of requests about airline flights. In spite of our shortfalls in understanding. There remain many difficult problems yet to be solved before digital speech processing will be considered a mature science and technology. we have been able to create remarkable speech processing systems whose performance increases at a steady pace. stock price quotations. specialized help desks etc. linguistics.Conclusion We have attempted to provide the reader with a broad overview of the field of digital speech processing and to give some idea as to the remarkable progress that has been achieved over the past 4–5 decades. Our basic understanding of the human articulatory system and how the various muscular controls come together in the production of speech is rudimentary at best. and perception provide the tools for 180 .

As we increase our basic understanding of speech. the application systems will only improve and the pervasiveness of speech processing in our daily lives will increase dramatically. with the end result of improving the productivity in our work and home environments.Conclusion 181 building systems that work and can be used by the general public. .

Of course. His patience. Editor-in-Chief of Now Publisher’s Foundations and Trends in Signal Processing.Acknowledgments We wish to thank Professor Robert Gray. which greatly improved the final result. for inviting us to prepare this text. technical advice. and editorial skill were crucial at every stage of the writing. and Tom Quatieri for their detailed and perceptive comments. Yariv Ephraim. we are responsible for any weaknesses or inaccuracies that remain in this text. 182 . Luciana Ferrer. We also wish to thank Abeer Alwan. Sadaoki Furui.

” Proceedings of IEEE. “Recursive windowing for generating autocorrelation analysis for LPC analysis. S. 1062–1066. 247–254. B. April 1982. Speech Coding. S. Hanauer. [4] B. Barnwell III.” IEEE Transactions on Acoustics. 1996. R.” IEEE Transactions on Communications. vol.References [1] J. S. November 1977.” Bell System Technical Journal. [9] T. Schroeder. [8] T. Remde. “A new model of LPC exitation for producing natural-sounding speech at low bit rates. and Signal Processing. and C. 4. H. S. pp. “Improved quantizer for adaptive predictive coding of speech signals at low bit rates. R. S. K. Atal and M. R. pp. 614–617. [6] B. pp. pp. vol. Barnwell III. B. Speech. [3] B. no. pp. pp. R. [2] B. 50. Nayebi. vol. Atal. vol. 65. October 1970. ASSP-27.” Proceedings of IEEE ICASSP. Rabiner. vol. 535–538. 600–614.” Journal of the Acoustical Society of America. no. Schroeder. [5] B. “A unified theory of short-time spectrum analysis and synthesis. Schroeder. pp. Richardson. 1971. S. Atal and J. 1558–1564. John Wiley and Sons. Atal and S. June 1979. pp. L. Allen and L. “Predictive coding of speech at low bit rates. Atal and M. 11.” IEEE Transactions on Acoustics. COM-30. 561–580. “Speech analysis and synthesis by linear prediction of the speech wave. B. April 1980. “Predictive coding of speech signals and subjective error criterion. October 1981. 183 . Atal and M. 5. 1973–1986. A Computer Laboratory Textbook. vol. “Adaptive predictive coding of speech signals. no. ASSP-29.” Proceedings of ICASSP. 1982. [7] B. Speech and Signal Processing. 49.

pseudo-autocovariance. no. Shoham. A. and A. New York: John Wiley and Sons. A. Garten. “Interaction of units in a unit selection database. vol. P. E. “A maximization technique occurring in the statistical analysis of probabilistic functions of Markov chains. 1986.” Proceedings of International Conference on Acoustics. 384–387. 2. R..” Bell System Technical Journal. Weiss.). E. Jayant. A. March 1992. K. Seshadri. H. 27. R. May 1989. [14] M. E. K. vol. [20] F.. 6. “Diphone synthesis using an overlap-add technique for speech waveform concatenation. Tukey. cross-cepstrum. vol.” Proceedings of IEEE ICASSP. no. 446–472. 2. and J. pp. 1972. Bresch. Inc.” Proceedings of ICASSP. Tremain. [19] J.” Inequalities. Nayak. Hungary. “Synchornized and noiserobust audio recordings during realtime MRI scans. Coker. N. Introduction to Wavelets and Wavelet Transforms. Gopinath. Bennett. Beutnagel. 1998. “Diphone synthesis using unit selection. November 1998. Conkie. “A model of articulatory dynamics and control. pp.” in Proceedings of the Symposium on Time Series Analysis. vol. Crochiere. [23] R. [25] R. W.” Proceedings of Eurospeech ’99. vol. Soules. September 1999. Campbell Jr. Rosenblatt. pp. 3. Multirate Digital Signal Processing.” Annals in Mathematical Statistics. and T. Schafer. V. M.” Bell System Technical Journal. 1976. Narayanan. vol. “Spectra of quantized signals. “Performance evaluation of analysis-bysynthesis homomorphic vocoders. vol. H. Petri.” Third Speech Synthesis Workshop. [12] W. 1069–1085. Budapest. 2015–2018. Quackenbush. 452–459. G. S. 1983. and S. pp. 41. “New directions in subband coding. 391–409. H. pp. E. October 2006. [24] R. 8. Syrdal. 55. Charpentier and M. [21] J. P. Kabal. [13] M. Rabiner. Beutnatel and A. Welch.184 References [10] L. pp.. [16] B.” Proceedings of IEEE. L. J. L. Stella. “Digital coding of speech in subbands. October 1976. 1791–1794. Mermelstein. 1970. vol. and N. pp. Y. R. Gay. Cox. . 117–120. Baum. Bogert. V.” IEEE Journal of Selected Areas in Communications. [17] E. Conkie. Nielsen. E.. pp. pp. [11] L. and P. no. P. [22] C. Australia. [18] C.” Journal of the Acoustical Society of America. 1–8. Speech and Signal Processing. G. 164–171. pp. Prentice-Hall Inc. ed. vol. and J. 2. and saphe cracking. Burrus and R. 1963. Chung and R. “An inequality and associated maximization technique in statistical estimation for probabilistic functions of Markov processes. T. Webber. Prentice-Hall Inc. (M. 735–738. Baum. and N. Healy. S. “Efficient computation and encoding of the multipulse excitation for LPC. C. March 1984. “An expandable errorpretected 4800 bps CELP coder. “The quefrency alanysis of times series for echos: Cepstrum. S. [15] M. Berouti. pp. 4.” Proceedings of ICASSP. 64. July 1948. W. 120. J. Crochiere and L. Jenolan Caes. Flanagan. S. February 1988.

R. Pinson. July 29. Springer-Verlag. C. pp. 122–126. Holliman. [37] H. Speech and Signal Processing. 3. Godfrey. vol. vol. no. “Comparison of parametric representations for monosyllabic word recognition in continuously spoken sentences.” Philips Research Reports. Speech Analysis. Fant. [36] J. Synthesis and Perception. pp. Dudley. and K.” Proceedings of ICASSP 1992. 1939. [33] J. L. Freeman Company. no. Shipley. 28. [27] S. B. 268– 278. measurement and calculation. pp. October 1933. Ghitza. [42] J.). “Speaker independent isolated word recognition using dynamic features of speech spectrum. pp. Coker. “Synthetic voices for computers. 442–448. October 1970. L. Ferguson.361. vol. Gold and L. Forney.” Hidden Markov Models for Speech. Rabiner. Flanagan. Ishizaka. 1991. “The vocoder. [38] G. Davis and P. 1993. no. pp. [34] J. vol. Speech. Furui. 2. [35] J. pp. 2. 2.605. pp. L. [28] F. NY: Marcel Dekker. Furui. “Delta modulation — a new method of PCM transmission using the 1-unit code. Princeton: Institute for Defense Analyses. . deJager. Umeda. Flanagan. The Hague: Mouton & Co. W. pp. K. “Differential quantization of communication signals. no. Rabiner. ASSP29. eds. Schafer. Flanagan. pp. Patent 2.” U. 7. Mermelstein. 442–466. 1952.” Bell Labs Record.” Bell System Technical Journal. 1. 485–506. “SWITCHBOARD: Telephone Speech corpus for research and development. 2nd Edition.” in Advances in Speech Signal Processing. [30] H. W. L. R. 1980. 2. 82– 108. 453– 485. J. L.” Journal of Acoustical Society of America. [32] G. Cutler.. pp.” IEEE Proceedings. [43] B. vol. E. 17. Netherlands: Kluwer Academic Publishers. Sondhi. 1972. Fletcher and W. pp. “Cepstral analysis technique for automatic speaker verification.” IEEE Transactions on Acoustics. March 1973. 1992. 1970.S. December 1952. Munson. 52–59. An Introduction to Text-to-Speech Synthesis. February 1986. 22–45. D.” IEEE Transactions on Acoustics. C. “Parallel processing techniques for estimating pitch period of speech in the time domain. vol. pt. “Hidden Markov Analysis: An Introduction.” IEEE Spectrum. “Loudness. “Synthesis of speech from a dynamic model of the vocal cords and vocal tract. Acoustic Theory of Speech Production. 357–366. “The Viterbi algorithm. Denes and E. 54. August 1980. J.” IEEE Transactions on Acoustics Speech. vol. August 1969. Signal Processing. 517–520. (S. March 1975. R.” Journal of Acoustical Society of America. April 1981. and J. pp. [41] O. The speech chain. H. B. and Signal Processing.References 185 [26] C. [39] S. Dutoit. H. no. 61. McDaniel. 46. 254–272. 1997. Furui and M. [40] S. [31] T. D. vol. ASSP-26. 1960. vol. and N. “Audiotry nerve representation as a basis for speech processing. C. its definition. 5. [29] P. Walter de Gruyter. N.

S. Saito. 4–28. 3. [53] F. pp. eds. [52] F. 51. Czech Republic. “Auditory modeling in automatic recognition of speech. Itakura and S. [47] J. “Line spectrum representation of linear predictive coefficients of speech signals. vol. pp. April 1984. 17–21. 1991. Dallas TX. Jayant and P. 57.” IEEE Signal Processing Magazine.” Journal of Acoustical Society of America. no. Jelinek. Wilpon. “Code modulation with digitally controlled companding for speech transmission. Gray.-W. vol. 1996. Jayant. [54] F. Statistical Methods for Speech Recognition. pp.” Bell System Technical Journal. pp. G. [59] F. Prague. Itakura. pp. Flanagan.” Bell System Technical Journal. 36–43. Spoken Language Processing. M. L.” in Proceedings of First European Conference on Signal Analysis and Prediction. “Adaptive quantization with one word memory. Roucos. R. [55] F. no. 1233–1268.). L. M. Greefkes and K. vol. 2001.” Proceedings of the Interactive Voice Technology for Telecommunications Applications (IVTTA).” Philips Technical Review. vol. L. Itakura and T.” Proceedings of ICASSP-96. Ishizaka and J. (S. 1996. C17–C20. 1. Riemens. Sachs. Parker. [57] N. 1972. Black. [50] A.” in Advances in Speech Signal Processing. [49] X. 1257– 1260. “Distance measure for speech recognition based on the smoothed group delay spectrum. [58] N. A. Atlanta. “How may I help you?. 2006. [45] R. [56] N. [60] F.” Proceedings of 6th International of Congress on Acoustics. 1970. pp. 335–353. and J. “Vector quantization. pp. 2. Acero. “Synthesis of voiced sounds from a twomass model of the vocal cords. Noll. p. [51] K. “Principles of lexical language modeling for speech recognition. Sondhi. Gorin. 373–376. s35(A). R. pp. S. Marcel Dekker. “A statistical method for estimation of speech spectral density and formant frequencies. Jelinek.” in Proceedings of ICASSP87. 57–60. “Adaptive delta modulation with a one-bit memory. pp.” Foundations and Trends in Communications and Information Theory. 321–342. April 1987. pp. Hermansky. 1984. 1. pp. 53-A. 651–699. PrenticeHall Inc. Mercer. A. 155–239. [48] H. Furui and M. no. B. vol. March 1970. [46] R. Umezaki. Gray. 1997. Cambridge: MIT Press. 1119–1144. Hunt and A. and H.” Bell System Technical Journal. . Prentice-Hall. “Unit selection in a concatenative speech synthesis system using a large speech database. M. S. 1970. 6. Hon. “Toeplitz and circulant matrices: A review. pp. M. and S. Huang. A. Saito. Itakura and S. pp. 535(a).” Electronics and Communications in Japan. Digital Coding of Waveforms. “Analysis-synthesis telephony based upon the maximum likelihood method. pp. September 1973. 1968. Jayant. 1997..186 References [44] A.

” IEEE Transactions on Acoustics. G. 1983. and M.” Proceedings of IEEE. J. [74] J. 2. [64] D. C.11. “A database for speaker-independent digit recognition. L. and J. 737–793. Lyon. ASSP-35. Markel. pp. [76] D. pp. 1986.” Bell System Technical Journal. 18. [70] Y. pp. “The SIFT algorithm for fundamental frequency estimation. pp. [77] J. Makhoul. 4. 2007. “The sound spectrograph. December 1978. and L. detection and compression in the cochlea. ASSP-34. “Linear prediction: A tutorial review. vol. 129–137. Gray. [66] W. 561–580. Leonard. and R.11. Lacey. [67] P.” IEEE Transactions on Audio and Electroacoustics. 42. 82. no. M. Deprettere. [62] B. vol. . pp. pp. vol. H. 1946. Speech Enhancement. G. W. Dunn. “Regular-pulse excitation: A nove approach to effective and efficient multipulse coding of speech.” Proceedings of ICASSP 1984. Y. [65] D. Linear Prediction of Speech. 1979. R. pp. H. no. H. no. Kroon. E. H. Paris. January 1980. pp. 5. Rabiner.” Journal of the Acoustical Society of America. Malah. Viswanathan. 2. 1976. July 1987. 84–95. “An algorithm for vector quantizer design. 27. [71] S. Huggins. R. and Signal Processing.” Journal of the Acoustical Society of America. “Maximum likelihood estimation for mixture multivariate stochastic observations of Markov chains. [69] S.” IEEE Transactions in Information Theory.” in Proceedings of IEEE International Conference on Acoustics. 121–133. 1985. [63] B. pp. Speech. Schwarz. Speech and Signal Processing. Wilpon. F. S. Theory and Practice. “A computational model of filtering.” Journal of the Acoustical Society of America. 64. pp. 6. [68] R. vol. “Time-domain algorithms for harmonic bandwidth reduction and time-scaling of pitch signals.” IEEE Transactions on Acoustics. Makhoul. vol. Koenig. L. May 1982.4. New York: SpringerVerlag. R. 1235–1249.” Journal of the Acoustical Society of America. Sluyter. pp. 28. 1984.” IEEE Transactions on Acoustics. E. “Software for a cascade/parallel formant synthesizer. H. H. 947–954. pp. 63. vol. December 1972. vol. A. 32. Markel and A.1–42. “Least square quantization in PCM. H. pp. 971–995. Rabiner. vol. Juang.” AT&T Technology Journal. vol. 367– 377. F. M. 64. Buzo. Levinson. Linde. vol. “Review of text-to-speech conversion for English. 1035–1074. and R. Loizou. AU-20. no.References 187 [61] B. Gray. “A mixed source model for speech compression and synthesis. Speech and Signal Processing. October 1986. March 1982. D. D. “Maximum likelihood estimation for multivariate mixture observations of Markov chains. [78] J. [72] P. “An introduction to the application of the theory of probabilistic functions of a Markov process to automatic speech recognition. 1054–1063. pp. France. 1980. “On the use of bandpass liftering in speehc recognition. Levinson. Speech and Signal Processing. V. vol. [73] R. Klatt. COM-28. CRC Press. no. 62. [75] J. 7. P. Juang. Sondhi. pp. vol. vol. Juang. September 1987. Lloyd. no. 307–309. K.” IEEE Transactions on Information Theory. and M. 19–49. M. and A.” IEEE Transactions on Communications. vol. Sondhi. 1975. Klatt. 67. E. 1577–1581. F.

vol. 9. August 1970. June 1968. Charpentier.” Journal of the Acoustical Society of America. V. no. July 1995. W. 5–6. “On the use of autocorrelation analysis for pitch detection. of Electronics. no. ASSP-25. “The 1994 benchmark tests for the ARPA spoken language program.” IRE Transactions on Information Theory. Buck.” IEEE Proceedings. Department of Electrical Engineering. vol. W. Technical Report No. pp. AU-16. 45. 8. R. “Nonlinear filtering of multiplied and convolved signals. “Cepstrum pitch determination. W. V. Oppenheim. Mohri. Rabiner. vol. vol. no. 9. pp. vol. Prentice Hall. MIT. [94] T. 293– 309.” IEEE Transactions on Audio and Electroacoustics. vol. Also: MIT Research Lab. 1999. [81] M.” Proceedings of IEEE. Oppenheim. no. “Homomorphic analysis of speech. “Minimum mean-squared error quantization in speech. 1264–1291. P. [89] A. 57–62. “Finite-state transducers in language and speech processing. Rabiner. 257–286. [85] A. V. “A speech analysis-synthesis system based on homomorphic filtering. 269–312. vol. vol. 23. no. 1964. [95] L. “Speech spectrograms using the fast Fourier transform. [80] A. March 1960. Schafer. 1973. February 1967. “A tutorial on hidden Markov models and selected applications in speech recognition. Pallett et al. 7. Cambridge. [96] L. pp. Oppenheim. Stockham Jr. August 1968. pp.” Proceedings of 1995 ARPA Human Language Technology Workshop. 7–12. April 1972. and J. R. vol. Com-20. 1989. 2002. [97] L. 432. 2. S. G. R. Oppenehim. [84] P. Prentice-Hall Inc. 1. 56.” IEEE Transactions on Communications. Moulines and F. [86] A. and Signal Processing. 2. “Pitch synchronous waveform processing techniques for text-to-speech synthesis using diphones.. Discrete-time speech signal processing.” MS Thesis. pp. and T. no. R. November 1975. V. 24–33. R. Rabiner. AU-17. 4. [88] A. [90] A. 1997. R. 293–309.. February 1977. H. 5–36. [83] A. 2. Quatieri. D. pp. 3. Portnoff. 1990. “A quasi-one-dimensional simulation for the time-varying vocal tract. no. 221–228. pp. 54. pp. pp. no. pp.” PhD dissertation. Schafer. 7–13. MIT. Noll. Speech. no. R. Paez and T. IT-6.” IEEE Transactions on Acoustics. [82] E.” Bell System Technical Journal. Schafer. vol. “Quantizing for minimum distortion. pp. vol. Discrete-Time Signal Processing. 242–250. Glisson.” Speech Communication.” Journal of the Acoustical Society of America. 225–230.188 References [79] J. March 1969. Oppenheim and R. [87] A. vol. Massachusetts. 1995. 1597–1614. vol. “A mixed excitation LPC vocoder model for low bit rate speech coding. “Superposition in a class of nonlinear systems. [91] M. 2. no. February 1969. 77. 1.” IEEE Spectrum. pp. . 41.” IEEE Transactions on Speech and Audio Processing. M. pp. “A comparative study of various schemes for speech encoding. Oppenheim. [92] D. V. F. Barnwell III.. “A model for synthesizing speech by rule. Max.” Computational Linguistics. V. [93] M.” IEEE Transactions on Audio and Electroacoustics. V. vol. 1965. McCree and T. pp. Noll.

Springer-Verlag. pp. 16. E. P. April 1986. 2. no. 35–43. no.” Proceedings of the International Conference on Acoustics. Flanagan. 1541–1558. Rabiner and B. “An algorithm for determining the endpoints of isolated utterances. vol. 1956. 2009. “System for automatic formant analysis of voiced speech. pp. Rosenfeld. [100] L.” Journal of Phonetics. Rabiner. Seneff. February 1975. “Homomorphic systems and cepstrum analysis of speech. vol. [112] M. “Two decades of statistical language modeling: Where do we go from here?. vol. Speech and Signal Processing. Sambur. February 1971. Dadson. Schroeter. [106] R. 679–682. “Code-excited linear prediction (CELP): High-quality speech at very low bit rates. PrenticeHall Inc. S. R. Schafer. 2000. “A vocoder for transmitting 10 kc/s speech over a 3. no. “Basic principles of speech synthesis. R. [105] A. 1985. D. Prentice-Hall Inc. E. R. May–June 1971. 7. vol. R. Rosenberg. vol. Sagisaka. C. Massachusetts. “A re-determination of the equal-loudness contours for pure tones. Schafer. 10.” PhD dissertation. R.References 189 [98] L. S. Rabiner and R. pp. “Rate-place and temporalplace representations of vowels in the auditory nerve and anteroventral cochlear nucleus. 1993. Robinson and R.” Journal of the Acoustical Society of America. “An introduction to hidden Markov models.” Journal of the Acoustical Society of America. 50. R. Also: MIT Research Laboratory of Electronics. R. Rose and T.” Proceedings of ICASSP ’86. 88. pp. vol. 937–940. L. Juang. “Speech synthesis by rule using an optimal selection of nonuniform synthesis units. 11. 466. (In preparation). 1985. [107] M.” Springer Handbook of Speech Processing and Communication. Cambridge. W. pp. “Computer synthesis of speech by concatenation of formant coded words. R. R.” Bell System Technical Journal. Rabiner and B. 822–828. Schroeder and B. Rabiner.” IEEE Proceedings. David. Schroeder and E. W. no. 458–465. Technical Report No. Schafer and L. vol. 166– 181. . 55–76. 47. 2. pp. R. W.” Springer Handbook of Speech Processing. H. [102] L. 54. Theory and Application of Digital Speech Processing. “The self excited vocoder — an alternate approach to toll quality at 4800 bps. [104] R. 297– 315. “A joint synchrony/mean-rate model of auditory speech processing. 43. 1969. MIT. H. Fundamentals of Speech Recognition. C. [103] D. Juang. vol. Sachs. Schafer. no.” Proceedings of IEEE ICASSP. pp. Schafer. Atal. Barnwell III. Young.” Journal of Phonetics. 1968. “Effect of glottal pulse shape on the quality of natural vowels. and E. B. W. 1988. 2007. [101] L. [109] R.” British Journal of Applied Physics. [111] R. February 1970.” Acustica. 1270–1278. [114] J. [110] R. Springer. 1960. pp. pp.” Bell System Technical Journal. and J. 16. Blackburn. vol. 37–53. pp. [108] Y. [113] M. 1988. 5. 2006.5 kc/s channel.. [115] S.” IEEE Signal Processing Magazine. vol.. C. pp. H. 4. [99] L. Rabiner and M. W. “Echo removal by discrete generalized linear filtering. 8. pp. 453–456. 1988. W.

“Instantaneous companding of quantized signals. 653–709. no. 63.” Proceedings of IEEE. Audio Signal Processing and Coding. Vaidyanathan. T. [117] G. 329. N. Ingebretsen.” IEEE Signal Processing Magazine. 15–24. Russell. [120] A. Tribolet. Speech. Ward. S. 27– 42. Nguyen. 2. 35. 53. ASSP-25. April 1967. pp. W. University of Illinois Press. 678– 692. Zwicker and H. “The relation of pitch to frequency. Fastl. and J. Fox. pp. J. no. T. Weaver. 702–710. 3. “Blind deconvolution through digital signal processing. 1. “Evaluation of the CMU ATIS system. and S. Viswanathan. vol. A. Linde. June 1975. pp. 20. Wiley Interscience. pp. Viswanathan and J. Stevens. R. and R. vol. C. 1940. M.” IEEE Transactions on Acoustical. vol. [123] L. pp. Makhoul. IT-13. M. vol. December 1975. Burrus. no. Makhoul. January 1993. November 2003. Soong and B.” IEEE Transactions on Acoustics. May 1957. “Voice-excited LPC coders for 9. [128] C. “A new phase unwrapping algorithm. 4. [122] S. “Quantization properties of transmission parameters in linear predictive systems. Tohkura. Psycho-acoustics. pp. Multirate Systems and Filter Banks.-H. Magill. pp. Speech. Atti. p. and Signal Processing. Springer-Verlag. April 1982. [131] V.” Bell System Technical Journal. J. Gray. April 1977. K. [119] F. Wellesley.190 References [116] C. 1466–1474. M. 101–105. 558–561. 6.” IEEE transactions on Communications. 2nd Edition. MA: Wellesley-Cambridge Press. 36. vol.. T. 1990.” IEEE Transactions on Communications. [125] G. 3. Un and D. Shannon and W. 1996. 1949. vol.” Proceedings of DARPA Speech and Natural Language Workshop. no. April 1975.” American Journal of Psychology. “Error bounds for convolutional codes and an asymptotically optimal decoding Aalgorithm. and Signal Processing. pp. [129] P.” IEEE Transactions on Speech and Audio Processing. pp. and V. 260–269. pp. “Factoring very-highdegree polynomials. Juang. February 1991. [126] Y.. [132] A. vol. Prentice-Hall Inc. Volkman. 170–177. [121] K. vol. . 1. “The design of trellis waveform coders. pp. Cannon. “The residual-excited linear prediction vocoder with transmission rate below 9. Stewart. Wavelets and Filter Banks. [127] J. Treitel. MIT Press. 1414–1422. W. C. vol. [134] E. Stockham Jr. P. 12. B. Acoustic Phonetics. Urbana. 2007. K. no. [133] W. E. vol. Spanias.6 kbits/s. COM-30. COM-23. The Mathematical Theory of Communication. ASSP-23.” IEEE Transactions on Information Theory. April 1979.” vol. Painter. Sitton.6 kbps speech transmission. Smith. “A weighted cepstral distance measure for speech recognition. vol. no. S. G. [118] B. and Y.” IEEE Transactions on Acoustics. Stevens and J. 309–321. October 1987. Speech and Signal Processing. [124] T. [130] R. “Optimal quantization of LSP parameters. 1998. Strang and T. Viterbi. 1993. pp.

• Journal of the Acoustical Society of America. 191 . we provide the following list of journals and books as a guide for further study. Main publication of IEEE Signal Processing Society. A publication of the European Association for Signal Processing (EURASIP) and of the International Speech Communication Association (ISCA). • Speech Communication. • IEEE Transactions on Speech and Audio. Listing the books in chronological order of publication provides some perspective on the evolution of the field. In addition. Publication of IEEE Signal Processing Society that is focused on speech and audio processing. General publication of the American Institute of Physics. Papers on speech and hearing as well as other areas of acoustics.Supplemental References The specific references that comprise the Bibliography of this text are representative of the literature of the field of digital speech processing. Speech Processing Journals • IEEE Transactions on Signal Processing. Published by Elsevier.

). Prentice Hall Inc. N. Schafer. G. 1999. Deller. Schafer and J. Stevens. • Advances in Speech Signal Processing. Second Edition. Markel (eds. T. Gold and N. D. J.. Marcel Dekker. SpringerVerlag. Synthesis and Recognition. Schafer. 1995. John Wiley and Sons. • Speech Processing. Berlin. W. M. 1987.. Morgan. R. Markel and A. M. 1987. 2008. R. Prentice Hall Inc. Prentice Hall Inc. E. L. • Speech and Audio Signal Processing. W. M. • Speech Coding. J. Springer. New York. K. B. 2002. Speech Coding References • Digital Coding of Waveforms.). 1979. Classic Reissue. Rabiner and R.. P. 1984. H.. 1978. Furui and M. • Practical Approaches to Speech Coding. Prentice Hall Inc. A Dynamic and Optimization-Oriented Approach. Furui. Nayebi. • Digital Speech Processing. Sondhi. • Speech Communication. IEEE Press Selected Reprint Series. W. 2009. 1992. Noll. 2001. Wiley-IEEE Press. 1991. Quatieri. MIT Press. Jayant and P. • Discrete-Time Processing of Speech Signals. B. H. • Discrete-Time Speech Signal Processing. Marcel Dekker Inc. Berlin. L. and J. • Linear Prediction of Speech. Kluwer Academic Publishers.. 1996. S. L. Barnwell and K. S. Sondhi and Y Huang (eds. . Addison-Wesley.. 1972. Gersho and R. K. • Theory and Application of Digital Speech Processing. D. O’Shaughnessy. New York. Flanagan. • Springer Handbook of Speech Processing and Speech Communication. Papamichalis. Benesty. M. N. Second Edition.. John Wiley and Sons. F. O’Shaughnessy. 1976.192 Supplemental References General Speech Processing References • Speech Analysis. J. L. W. R. T. Kleijn and K. • Acoustic Phonetics. Springer Verlag. • Vector Quantization and Signal Compression. S.. P. J. Rabiner and R.. A. Gray. Elsevier. J. Human and Machine. D. 1998. • Digital Processing of Speech Signals. L. 2000. Synthesis and Perception. Prentice-Hall Inc. A Computer Laboratory Textbook. • Speech Coding and Synthesis. Paliwal. 2003. Jr. Deng and D. Hansen. • Speech Analysis. Jr. Marcel Dekker Inc. Gray. Proakis.

Prentice Hall Inc. P. • Spoken Language Processing. SpringerVerlag. S. Childers. Prentice Hall Inc. Morgan. Kluwer Academic Publishers. • Mathematical Models for Speech Technology. Sagisaka. 1996. R.. 1994. 2005. J. D.. • Foundations of Statistical Natural Language Processing. . Hirschberg (eds. Klatt. C. Paliwal (eds. W. Kluwer Academic Publisher. S. P. John Wiley and Sons. Alwan (eds. P. 1997. • Text-to-Speech Synthesis. Hunnicutt and D. Hon. 2000. Olive. Rabiner and B.. M.. 2008. D. • An Introduction to Text-to-Speech Synthesis. J. • Connectionist Speech Recognition-A Hybrid Approach. Levinson. Speech Recognition and Natural Language Processing • Fundamentals of Speech Recognition. Cambridge University Press. Lee. Second Edition. Sproat. • Progress in Speech Synthesis. • Speech Coding Algorithms. C. Jurafsky and J. J. Campbell and N. Coleman. 1999. Springer-Verlag. A. Greenwood and J. John Wiley and Sons. 2003. • Statistical Methods for Speech Recognition. N. Juang. F. Kluwer Academic Publishers. • Automatic Speech and Speaker Recognition. Acero and H. H. Riek. Manning and H. S. C. • Acoustics of American English. L. 1987. 2004. VanSanten. 1996. A.-W. 1999. Jelinek. J. • Computing Prosody. Schutze. • Digital Speech: Coding for Low Bit Rate Communication Systems.Supplemental References 193 • A Practical Handbook of Speech Coders. Bourlard and N. Higuchi. T. 2000. 1996. R. Allen. Martin. • Speech and Language Processing. MIT Press. Y. John Wiley and Sons. 1993. Taylor. Narayanan and A. Prentice Hall Inc. 2000.). Cambridge University Press. H. F. 2004. Olive and J. A. E. X. Soong and K. Huang.). Dutoit. K. CRC Press. Chu. W. Goldberg and L. • Text To Speech Synthesis: New Paradigms and Advances. D. • Speech Processing and Synthesis Toolboxes. Prentice Hall Inc. H. A. K. H. Speech Synthesis • From Text to Speech.). John Wiley and Sons. 1998. MIT Press. Springer-Verlag. 1993. Kondoz. R.

• Audio Signal Processing and Coding. . H. Loizou. Brandenburg (eds.194 Supplemental References Speech Enhancement • Digital Speech Transmission. 2007. Atti. T.. 2007. Theory and Practice. John Wiley and Sons. Kahrs and K. Spanias. Vary and R. C. P. 2006. A. John Wiley and Sons. CRC Press. P. Ltd. 1998. • Speech Enhancement. Audio Processing • Applications of Digital Signal Processing to Audio and Acoustics. Enhancement.). Coding and Error Concealment. Martin. Kluwer Academic Publishers. Painter and V.

Sign up to vote on this title
UsefulNot useful