The Color of Sound: A Theoretical Study
in Musical Timbre
Wayne Slawson
Introduction
Musical timbre (or instrumental color, Klangfarbe, sound
quality —there are many terms each with its own shade of
meaning) has resisted the efforts of music theorists. Writers
such as Pierre Schaeffer (1966), Cogan and Escot (1976) and
Robert Erickson (1975) have contributed valuable insights.
However, we do not yet have a theory that is consistent with
what is known about the auditory system, that describes musical
practice well, and that defines a rich set of operations through
which new music can be structured.
The difficulty lies partly with the term itelf. Timbre has been
used in too many contexts to mean too many different things. If
it is used to refer to the identifying sound of a musical instru-
ment, as it is in the original French usage, it would seem that we
would have to conclude that an instrument—say a clarinet—has
a single timbre. But of course instruments have different "qual-
ities" or "colors" in their different registers—the clarinet even
has names for its various registers. The American Standards . I am treating sound color as an abstract auditory phenome-
Association definition is flawed in the opposite way; it assigns
the term to any difference between sounds of the same loudness,
pitch, and duration. Theorists who take as their subject this
catch-all meaning of the term—notably Schaeffer and
Erickson—are faced with an enormous range of phenomena.
Their response, to catalog what they consider significant types
of timbres, is about the only possible approach.
Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015
For my own contribution to the subject, I have chosen to limit
the range of my claims to a subset of timbre I call sound color.
Sound color as I am using the term is a psychoacoustic
"attribute" or set of attributes of sound, joining such familiar
attributes as pitch and loudness. Sound color does not necessar-
ily refer to musical instruments; it is, rather, an abstract property
of auditory sensation. By definition it has no temporal aspect.
Sounds may vary in color over time, but the variation in a sound
is not itself a color. Thus, for example, the "graininess" of a
sound, its degree of vibrato, and the characteristics of its attack
are not aspects of sound color.
The term is not meant in any synesthetic sense, but the
analogy with visual color is a good one. Visual color is a
psychological attribute. It is multidimensional and it has no
temporal aspect. Lights that appear to vary in color are not said
to have a single color, but a succession of different colors.
non, but the concept grows out of a model of the way certain
kinds of natural sounds arise. This model is called the source-
filter model of sound production.

The Source-Filter Model of Sound Production
According to this model, sound is produced when mechanical
or acoustic energy excites vibration in an object or cavity
(Huggins, 1952; Fant, 1960) . The excitation is the "source" in
the model and it is affected only negligibly by the object or
cavity that it excites (see Figure 1) . Typically the temporal
features of the sound—the time of its occurrence, its envelope,
and its frequencies and overall intensity—can be attributed to
the source. The spectrum of the source might be as illustrated in
Figure 1. The influence of the object or cavity on the sound is
represented by the spectrum envelope — a characteristic shape
(like the dotted curve in Figure 1) that is imposed on the
spectrum of the excitation, modifying the intensities but not the
frequencies of the components of the excitation. The spectrum
envelope is the "filter" in the source-filter model. Within broad
limits, the excitation can be of almost any form—noise, a
low-pitched sound, a higher pitched sound, a series of
impulses—in each case the spectrum envelope "colors" the
source spectrum.
Figure 1. The source-filter model of sound production
SOURCE FILTER SOUND
111
AM PLI
Il i
FREQ.
- 11 ' br-f-i
SOURCE SPECTRUM RESULTANT
SPECTRUM ENVELOPE SPECTRUM
The Color of Sound 133
There are many examples of such sounds. The main require-
ment is simply that the excitation be independent of the sound-
ing object. Natural sounds in our environment often meet this
requirement quite well. The model is particularly good at de-
scribing the production of speech. In vowels the excitation is
generated by the vocal folds in the larynx; the filter is the throat
and mouth. The tension of the vocal folds and the volume of air
Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015
we force through them control the pitch and intensity of the
vowel. The shape of the vocal tract controls the vowel quality or
color. As we all know, the excitation and the "filter" in vowels
are independent. We can utter a soft, low-pitched [a] or a loud,
high-pitched [a] or we can utter an [a] and an [e] at the same
moderate intensity and pitch. The source-filter model also fits a
very common means of sound production in the electronic
studio: the, "treatment" of a complex sound by an independ-
ently controlled filter. Seldom indeed do we encounter a piece of
electronic music that makes no significant use of this technique.
Musical instruments, by and large, are not very well de-
scribed by the model, because in most of them the excitation and
the filter are "strongly coupled." The resonances in the air
column or string—the "filter" —react back on the buzzing lips,
the vibrating reed, or the moving bow hairs—the source. This
"strong coupling" causes the source to be driven at one of the
resonance frequencies of the filter. In these cases the filter
controls the pitch and indirectly the loudness of the sound. The
situation is complex, however, for there are additional reso-
nances in musical instruments that act much like those in the
human vocal tract and in the electronic studio. In fact, although
the source-filter model is not intended to apply to musical
instruments, it is consistent with certain of the ideas of scholars
like Cogan and Escot (1976) who theorize about the color of
instrumental combinations.
Keeping in mind these rather elaborate fair warnings about
what sound color is and is not about, let us turn to the central
questions that a theory of sound color must address. They are
concerned with invariances. In order from the simple and gen-
134 Music Theory Spectrum

eral to the more complex and specifically musical, these ques- Figure 2. Alternative transformations of the spectrum of a sound. When
tions are: the fundamental frequency of a sound (X) is raised by an octave, the
spectrum envelope can be held constant (A) or the relative intensities of
1. How do we hold color invariant when other parameters of the partials can be held constant (B).
sound are changed?
2. How do we hold one aspect of sound color invariant when
other aspects are changed? This is equivalent to asking for a

Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015

description of the dimensions of color.
3. What operations can we define that preserve invariant
relations among different sound colors? Or more generally, how X
can color be structured in a musical context?
I 2 3 4 5 6 7 8 9 to
I
- •
^ I
FREQUENCY

Invariance of Sound Color

The answer to the first question is somewhat deceptively A
simple. It is the following rule: To keep sound color constant as
other parameters (such as pitch and loudness) are changed,
keep the spectrum envelope constant. Figure 2 illustrates this
claim and a popular alternative. When we raise the pitch of a
sound (X in Figure 2) by an octave, we can either keep the "hills
and valleys" in the sp e ctrum at the same places (alternative A),
or we can keep the relative intensities of the partials constant
3 4
(alternative B). The first alternative is the result of following the
rule I have just stated. The second alternative is the result of such
operations as doubling the speed of a tape recorder or of repro-
ducing the shape of a single cycle of the waveform in half the
time.
Helmholtz (1885) seems, in one place (pp. 19-25), to favor with the "brightening" of recordings played back too fast and of
the latter alternative as a means of keeping Klangfarbe constant. the invariance of vowel qualities supports the first over the
His endorsement is by no means unequivocal, however, and in second alternative as well. A few studies suggest even that
other contexts he can be interpreted as favoring the first alterna- certain musical instruments can be identified on the basis of their
tive (e.g. , p. 128) . A number of scholars since Helmholtz's invariant resonances (e.g., Grey, 1978). Just how the auditory
pioneering work have also tended to favor the first alternative system handles these sensory invariances is not at all well
(e.g. , Stumpf, 1926; Chiba and Kajiyama, 1941) . understood by physiologists. But be that as it may, psychologi-
Recently a number of psychoacoustic studies (Slawson, cal evidence, our everyday experiences, and certain kinds of
1968; Plomp and Steeneken, 1971) have strongly favored the invariances in the sounds of musical instruments, all support the
fixed spectrum envelope paradigm. Our everyday experience first basic claim of the theory.
 The Color of Sound 135

The Dimensions of Sound Color Figure 3. The sound color space. The dimension of OPENNESS varies
roughly with the frequency of the first resonance; ACUTENESS, with the
With the "fixed spectrum envelope rule" we can preserve a frequency of the second. The middle of the space (at the [ne] sound) is
particular sound color, but can we say something more about the the maximally LAX sound; around the periphery are the least LAX sounds.
color itself? In what ways does it differ from other colors? What
kind of world does it reside in? These questions lead, of course,
to the second of the central questions I have posed concerning

Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015

the dimensions of sound color.
Of the several dimensions of sound color, I shall discuss three
that seem to be the most salient and musically useful. They are N 2.5
called LAXNESS, OPENNESS, and ACUTENESS. In order to explain what
these dimensions are I shall have to interject a brief tutorial here
about the acoustics of vowels. „ „ 2
^^ n
As illustrated in Figure 3, vowel sounds can be associated bit bet
with certain combinations of frequencies of the lowest two
41,NbiN,N,
resonances in their spectrum envelopes. The "long" vowels are
located around the periphery of the space. Moving toward the
0# 3

center of the space we encounter the "short" vowels, which

occur in words like "bit," "bet,"and "but."In the center is
f e
buf
the neutral vowel, the color to which most vowels deteriorate, in
"
a
English and many other languages, when they are in unstressed
syllables .
The dimensions of color are defined in terms of these spec-
trum envelope peaks as well. As a first approximation, OPENNESS
varies with the frequency of the first resonance; ACUTENESS with GUN
/. 0
the frequency of the second. Thus when both resonances are low ^

in frequency, we have the non-AcuTE, non-oPEN vowel, [u]. A ^

high second resonance and a low first resonance produces an a

0.8-
[i]-like, ACUTE, non-OPEN sound. A high first resonance with a LL.1
cfi 0
low second resonances produces [aw] as in "maudlin" -OPEN
and non-ACUTE. When both resonances are high, we get the OPEN U. 6 -
u
and ACUTE [ae] as in "bad. " I should point out here that the name
of the OPENNESS dimension is derived from the shapes of tubes 0.2 0.4 a.6 o.8 I.O
that produce OPEN sounds. These tubes have no significant nar-
rowing; they are relatively "open." Tubes associated with FIRST RESONANCE REQUENCY(KHZ)
non-OPEN sounds, on the other hand, all have at least one region
136 Music Theory Spectrum
of small cross-sectional area. In the vowel [u], the narrowing is
between the lips; in [i], it is between the tongue and the hard
palette. The OPEN vowels [aw], [a], and [ae] have no such
narrowings. The color in the middle of the space, corresponding
to the neutral vowel, is maximally LAX. The colors located
around the middle are less lax and the colors in the periphery of
the space are least LAX.
I can be more specific about the dimensions by drawing equal
value contours for each of them. All configurations of the first
two resonances that fall on a single contour have the same value
with respect to the dimension in question. To locate the contours
exactly would demand a good deal of psychoacoustic research,
but I can at least suggest what their approximate shapes will be.
In Figure 4 are plotted hypothetical equal-LAXNESS, equal-
OPENNESS, and equal-ACUTENESS contours.
Notice that the contours tell us how to hold one aspect of
sound color constant while varying other aspects. Suppose for
example, we want to hold LAXNESS invariant in the face of
changes in OPENNESS and ACUTENESS. We would select one of the
equal-LAXNESS contours and then move along it with the proper
variations in the resonance frequencies. Similarly we can move
along an equal-oPENNESs contour by changing mostly the fre-
quency of the second resonance while holding the first almost
constant. In general this kind of transformation produces
changes in ACUTENESS and LAXNESS but not in OPENNESS.
Now, these dimensions are not plucked full-blown from the
air. They are extensions of what is known as the "distinctive
feature" theory in phonetics (Jakobson et al. , 1951) . This the-
ory is held in high regard by linguists because, among other
things, it provides a basis for understanding the properties that
all languages have in common (Chomsky and Halle, 1968).
What I have done is to treat three of the most fundamental of the
"features" as continuous dimensions and to apply them to
sound in general. While this may seem an audacious step, it is
not a terribly large one. There are grounds both rational and
empirical to hold that OPENNESS and ACUTENESS and possibly
LAXNESS can be applied to non-speech sounds. But the most
compelling evidence is musical. It is necessary, I believe, to
appeal to something like the dimensions I have postulated to
describe adequately the ways in which sound color has been
structured in a number of electronic compositions.
One example is both brief and particularly striking. The
introduction of Milton Babbitt's Ensembles for Synthesizer
Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015
consists of four phases each ending in a complex, nearly static
sound. These four static sounds serve to expose the color world
of the piece. (They are most easily compared if they are isolated
and juxtaposed.) The first sound is roughly [i]-like, the second
[aw]-like, the third [ae]-like, and the fourth [u]-like. But the
sounds are more than simple vowels. They are complex mix-
tures of sounds that include what I have called LAx sounds as
well. Each sound seems to fill a quadrant of the two-dimensional
space whose axes are OPENNESS and ACUTENESS. Notice that it
would not be possible to fit them into a single high/low,
"brightness" dimension. The two-dimensional space is re-
quired. This is only one of a number of examples from works by
a varety of composers in which one or the other of the two
dimensions of ACUTENESS and OPENNESS are exploited in signifi-
cant ways.
Operations on Sound Color
In the pieces I have studied, I have found no unequivocal use
of the LAXNESS dimension by itself. It is of great importance
theoretically, however, for the maximally LAx point is a singu-
larity upon which a number of operations on sound colors can be
defined. But LAXNESS appears to have musical potential in its
own right for it too possesses a degree of sensory naturalness.
This is best demonstrated in terms of the first operation I shall
describe: the sound color analogy to transposition.
 The Color of Sound 137

Figure 4. The dimensions of sound color. Loci of equal value for each
of the dimensions of LAXNESS, OPENNESS and ACUTENESS are plotted in terms
of the frequencies of the first two resonances. The approximate posi-
tions of some representative vowels are indicated.

EQUAL LAXNESS
C 0 NTOURS

Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015

2.s

2. 0

%•5
2
o
V
^
Lu
c/)
lt. u
0
? 1.0
v
o.e
h

u_ 0. 6

0.2 o.4 o.lo o.S 1. 0

FREQUENCY OF Fl KST
RE SON AN c.E (KNZ)

EQUAL OP EN NESS EQUAL ACUTENESS

CONTOURS CONTOURS
.^

2S -

""

2.0 e

1.5 1.5

CL
i
1 ,0 1.0

0,8 0.g

0.6 0.6
, ^

0.2 0. 4 o.b o.8 i.o 0.2 0.4 o. 6 0.81 .0

1
138 Music Theory Spectrum

Transposition of Sound Color Unlike transposition in the pitch domain, color transposition
is not closed in the algebraic sense. If for example we had
Transposition of sound color is simply a shift of the resonance
attempted to transpose the [u] sound in the direction of negative
peaks in a direction perpendicular to the equal-value contours of
ACUTENESS, we would have required physically unrealizable
Figure 4. It is defined for each of the dimensions and consists of
negative frequencies for the second resonance. Similarly, a
adding or subtracting some constant from the value of a color on
negative transposition constant for OPENNESS on the [u] sound
a particular dimension. Suppose we have a sequence of sound
would produce an impossible value for the first resonance.

Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015

colors that we can associate with the vowels [u], [ne], [o] (see
Clearly, transposition is not defined for certain combinations of
Figure 5). A transposition in ACUTENESS of this sequence would
transposition constants and color values. The situation is a
produce something like the sequence [oe], [e], [ne]. A subse-
complex one in which questions of the metric of the space and an
quent transposition in OPENNESS of the ACUTENESS transposed
additional dimension, not discussed here, are involved.
sequence would produce the sequence [ne], [ae], [a].
One theoretically "clean" solution is to define another kind
of transposition —call it translation —in which the generation of
impermissible colors is avoided by a kind of "wrap-around. " A
negative constant of translation in ACUTENESS on the color asso-
Figure 5. Transpositions in ACUTENESS and OPENNESS
ciated with [u] would result in, not some kind of impossibly
"super"-[u], but an [ii-like sound. By the same token, a posi-
tive translation in ACUTENESS of an [i] sound would result in
[u] . Unfortunately, there is no psychoacoustic justification for
this more theoretically satisfying operation. Regularities based
acuteness
t+ness -^ on it would have to depend on a learning process; a dependency,
^h ?e
npermess incidently, that is not without precedent in other areas of music.
b-dnsposi fiion
Transposition in LAXNESS is a somewhat different kind of
oe ime .,. operation. As in the case of the other dimensions, certain trans-
Ji positions are ruled out. Since maximal LAXNESS is the neutral
position in the center of the sound color space, LAXNESS values
Ari9ina^ seg uev►ce
beyond that maximal value are undefined. Similarly, sounds can
be only so non-LAX. Within the permissible range of transposi-
tion in this dimension, however, we have what I believe are
0 musically compelling possibilities. I have synthesized a se-
0.6
quence of four sound colors repeated several times, each time at
o z o.4 c. h o. g 1.0
a different pitch. The sequence begins at the neutral, maximally
FIRS T RESON A NCE O<H1)
LAX position and the four colors are undifferentiated. Then a
transposition in LAXNESS is imposed and the sequence is heard to
become gradually less LAX. Then from the greatest non-LAX
positions, the sequence is transposed once more with respect to

LAXNESS in the opposite direction, returning step by step to the
undifferentiated neutral position. The shifts in the resonance
frequencies that accomplish these two transpositions are quite
complex. The sound of the operation, to my ears at least, is quite
simple and natural. I have resynthesized the sequence with a
complex, frequency modulated source and with a noise source
with no deterioration in the perceived simplicity and naturalness
of the operations.
Inversion of Sound Color
The neutral position is a key to the definition of another
important transformation of sound color, inversion. This opera-
tion is best understood by visualizing a projection of the maxi-
mally lax point onto each of the other two dimensions. The loci
of points where this projection intersects the equal value con-
tours of one dimension defines arbitrary zero points on the other
dimension. Figure 6 illustrates this projection.
Figure 6. Axes of inversion projected from the neutral point
^
U. .. ,,
1
,/
^
ZERO ,
0.2 0.4 0.6 o'.g 1:0
FIRST RESONANCE (KHZ)
The Color of Sound 139
To invert a color with respect to one of the dimensions, we
simply negate the value of the color in terms of the zero point on
that dimension's axis. Thus the OPEN [a] sound inverted with
respect to OPENNESS becomes the non-oPEN [oe]. An ACUTENESS
inversion of the ACUTE [i]-like sound becomes an [uj-like sound.
Inversion with respect to OPENNESS of the same [i]-like color
results in the OPEN [ae]-like color. All colors have inverses with
Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015
respect to OPENNESS and ACUTENESS, so the inversion operation,
unlike transposition, produces closure in the space; inversion is
a group operation.
To illustrate inversion let us consider a sequence of colors we
may represent by the vowels [awl, [a], [o], [e], [u] (see Figure
7) . Inverting the sequence with respect to ACUTENESS produces
the sequence [ae], [a], [e], [o], [i]. Inversion with respect to
OPENNESS of the original sequence ([aw], [a], [o], [e], [u]) pro-
duces the sequence [u], [oe], [o], [e], [aw]. Figure 8 illustrates
the operation. Since the maximally LAX point defines the axis
about which inversion in OPENNESS and ACUTENESS takes place, no
inversion with respect to LAXNESS itself is permitted in the the-
ory.
Open Questions
One of the dimensions of sound color that I have omitted from
this discussion provides a means of extending the realm of sound
color beyond the range of vowel-like colors. I think I know how
to treat that dimension, but I have changed my mind about it at
least once in recent months and it requires a bit more study.
Sound color mixture is another topic that I am not sure how to
deal with. I suspect something more interesting than treating
two or more simultaneous colors as some kind of sum of the
individual colors is possible, but I have not yet found anything
that is really satisfactory.
Needless to say, many, many empirical questions remain
unanswered. Where exactly are the equal-OPENNESS, equal-
ACUTENESS, and equal-LAXNESS contours? Can the counter intui-
tive operation of translation produce psychologically realistic
140 Music Theory Spectrum
Figure 7. Inversion with respect to ACUTENESS
0.2 0. 4 0.6 0.8 1.0
FIRST RESO N ANCE (KHZ)
invariances? Can we learn to hear as invariants the transforma-
tions of that operation?
The empirical question that is most important, of course, is
whether or not composers can apply the theory successfully.
Because composers have already controlled color in ways that
are consistent with the theory there are some grounds for opti-
mism. My own experience with a piece in progress leads me to
suspect that the most troublesome problems will have to do with
the integration of sound color with the more traditional musical
elements. I suppose it is not necessary to add that I find such
problems very interesting and provocative. I hope other com-
posers will be attracted to them as well.
I am indebted to Robert Morris and John Peel for suggestions that have
improved the theory and this paper significantly. Robert Cogan's objection to
the counter intuitive name I had adopted for the OPENNESS dimension guided
Figure 8. Inversion with respect to OPENNESS
Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015
02 0 4 0.6 0.8 1.0
F IRST RESONANCE e<HZ)
References
Chiba, T. and M. Kajiyama. The Vowel: Its Nature and Structure. Tokyo:
Tokyo-Kaiseikan, 1941.
Chomsky, N. and M. Halle. The Sound Pattern of English. New York: Harper
and Row, 1968.
Cogan, R. and P. Escot. Sonic Design: The Nature of Sound and Music.
Englewood Cliffs, N. J.: Prentice-Hall, 1976.
Erickson, R. Sound Structure in Music. Berkeley: University of California
Press, 1975.
me to my present choice. The work reported in this paper was supported by a
Fellowship from the American Council of Learned Societies, the National
Endowment for.the Arts—a federal agency, the A. W. Mellon Educational and
Charitable Trust, the Westinghouse Corporation, and the University of Pitts-
burgh.
 The Color of Sound 141

Fant, G. Acoustic Theory of Speech Production. The Hague: Mouton, 1960.

Grey, J. M. "Timbre Discrimination in Musical Patterns." J. Acoust. Soc.
Am., 64 (1978), 467-72.
Helmholtz, H. von. On the Sensations of Tone. London: Longmans 1885.
The original English translation has been reprinted by Dover, 1954.
Huggins, W. H. "A Phase Principle for Complex-Frequency Analysis and Its
Implications in Auditory Theory." J. Acoust. Soc. Am., 24 (1952), 582-89.
Jakobson, R., G. Fant, and M. Halle. Preliminaries to Speech Analysis: The

Downloaded from http://mts.oxfordjournals.org/ at UQ Library on June 19, 2015

Distinctive Features and Their Correlates. Cambridge, Mass.: MIT Press,
1951
Plomp, R., and H. J. Steeneken. "Place Dependence of Timbre in Reverber-
ant Sound Fields." Acustica, 28 (1971), 50-59.
Schaeffer, P. Traitē des objets musicaux. Paris: Editions du Seuil, 1966.
Slawson, A. W. "Vowel Quality and Musical Timbre as Functions of Spec-
trum Envelope and Fundamental Frequency." J. Acoust. Soc. Am., 43 (1968),
87-101.
Stumpf, C. Die Sprachlaute. Berlin: Springer, 1926.