Professional Documents
Culture Documents
Download textbook Mathematical And Engineering Methods In Computer Science 9Th International Doctoral Workshop Memics 2014 Telc Czech Republic October 17 19 2014 Revised Selected Papers 1St Edition Petr Hlineny ebook all chapter pdf
Download textbook Mathematical And Engineering Methods In Computer Science 9Th International Doctoral Workshop Memics 2014 Telc Czech Republic October 17 19 2014 Revised Selected Papers 1St Edition Petr Hlineny ebook all chapter pdf
https://textbookfull.com/product/combinatorial-algorithms-25th-
international-workshop-iwoca-2014-duluth-mn-usa-
october-15-17-2014-revised-selected-papers-1st-edition-
kratochvil-jan/
https://textbookfull.com/product/digital-forensics-and-
watermarking-13th-international-workshop-iwdw-2014-taipei-taiwan-
october-1-4-2014-revised-selected-papers-1st-edition-yun-qing-
Engineering Multi Agent Systems Second International
Workshop EMAS 2014 Paris France May 5 6 2014 Revised
Selected Papers 1st Edition Fabiano Dalpiaz
https://textbookfull.com/product/engineering-multi-agent-systems-
second-international-workshop-emas-2014-paris-france-
may-5-6-2014-revised-selected-papers-1st-edition-fabiano-dalpiaz/
https://textbookfull.com/product/biometric-authentication-first-
international-workshop-biomet-2014-sofia-bulgaria-
june-23-24-2014-revised-selected-papers-1st-edition-virginio-
cantoni/
https://textbookfull.com/product/parameterized-and-exact-
computation-9th-international-symposium-ipec-2014-wroclaw-poland-
september-10-12-2014-revised-selected-papers-1st-edition-marek-
Petr Hlinený · Zdenek Dvorák
Jirí Jaroš · Jan Kofron · Jan Korenek
Petr Matula · Karel Pala (Eds.)
Mathematical and
LNCS 8934
Engineering Methods
in Computer Science
9th International Doctoral Workshop, MEMICS 2014
Telč, Czech Republic, October 17–19, 2014
Revised Selected Papers
123
Lecture Notes in Computer Science 8934
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen
Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zürich, Zürich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7408
Petr Hliněný · Zdeněk Dvořák
Jiří Jaroš · Jan Kofroň
Jan Kořenek · Petr Matula
Karel Pala (Eds.)
Mathematical and
Engineering Methods
in Computer Science
9th International Doctoral Workshop, MEMICS 2014
Telč, Czech Republic, October 17–19, 2014
Revised Selected Papers
ABC
Editors
Petr Hliněný Jan Kořenek
Masaryk University Brno University of Technology
Brno Brno
Czech Republic Czech Republic
Jan Kofroň
Charles University
Praha
Czech Republic
This volume contains the proceedings of the 9th Doctoral Workshop on Mathematical
and Engineering Methods in Computer Science (MEMICS 2014) held in Telč, Czech
Republic, during October 17–19, 2014.
The aim of the MEMICS workshop series is to provide an opportunity for PhD
students to present and discuss their work in an international environment. The scope
of MEMICS is broad and covers many fields of computer science and engineering.
In the year 2014, submissions were invited especially in the following (though not ex-
clusive) areas:
– Algorithms, logic, and games,
– High performance computing,
– Computer-aided analysis, verification, and testing,
– Hardware design and diagnostics,
– Computer graphics and image processing, and
– Artificial intelligence and natural language processing.
There were 28 submissions from PhD students from 10 countries. Each submis-
sion was thoroughly evaluated by at least four Program Committee members who also
provided extensive feedback to the authors. Out of these submissions, nine full papers
were selected for publication in these post-proceedings, and additional nine papers were
selected for local presentation at the workshop.
The highlights of the MEMICS 2014 program included six keynote lectures de-
livered by internationally recognized researchers from the aforementioned areas of
interest. The speakers were:
– Gianni Antichi from University of Cambridge who gave a talk on “An Open-Source
Hardware Approach for High Performance Low-Cost QoS Monitoring of VoIP
Traffic,”
– Derek Groen from University College London who gave a talk on
“High-Performance Multiscale Computing for Modelling Cerebrovascular Blood-
flow and Nanomaterials,”
– Jozef Ivanecký from European Media Laboratory who gave a talk on “Today’s
Challenges for Embedded ASR,”
– Daniel Lokshtanov from University of Bergen who gave a talk on “Tree Decompo-
sitions and Graph Algorithms,”
– Michael Tautschnig from Queen Mary University of London who gave a talk on
“Automating Software Analysis at Large Scale,” and
– Stefan Wörz from University of Heidelberg who gave a talk on “3D Model-Based
Segmentation of 3D Biomedical Images.”
The full papers of four of these keynote lectures are also included in the proceedings.
In addition to regular papers, MEMICS workshops traditionally invite PhD students
to submit a presentation of their recent research results that have already undergone a
VI Preface
rigorous peer-review process and have been presented at a high quality international
conference or published in a recognized journal. A total of 16 presentations out of
22 submissions from 6 countries were included into the MEMICS 2014 program.
The MEMICS tradition of best paper awards continued also in the year 2014. The
best contributed papers were selected during the workshop, taking into account their
scientific and technical contribution together with the quality of presentation. The 2014
awards went to the following papers:
– “LTL Model Checking of LLVM Bitcode with Symbolic Data” by Petr Bauch,
Vojtěch Havel, and Jiří Barnat; presented by Petr Bauch.
– “A New Concept in Advice Complexity of Job Shop Scheduling” by David Wehner;
presented by the author.
The two awards consisted of a diploma accompanied by a financial prize of 400 Euro
each. The prize money was donated by Red Hat Czech Republic and by Y Soft, two
of the MEMICS 2014 Industrial Sponsors.
The successful organization of MEMICS 2014 would not be possible without gen-
erous help and support from the organizing institutions: Brno University of Technology
and Masaryk University in Brno.
We thank the Program Committee members and the external reviewers for their
careful and constructive work. We thank the Organizing Committee members who
helped to create a unique and relaxed atmosphere which distinguishes MEMICS from
other computer science meetings. We also gratefully acknowledge the support of the
EasyChair system and the great cooperation with the Lecture Notes in Computer Sci-
ence team of Springer Verlag.
Workshop
The 9th Doctoral Workshop on Mathematical and Engineering Methods in Computer
Science (MEMICS 2014) was held in Telč, Czech Republic, during October 17–19,
2014. The workshop was hosted by the Telč University Centre of Masaryk University
(former Jesuits College), in a small and pleasant town of Telč inscribed in the UN-
ESCO List. The workshop was attended by 70 participants. More information about the
MEMICS workshop series is available at http://www.memics.cz/.
General Chair
Petr Hliněný Masaryk University, Czech Republic
Program Committee
Gianni Antichi University of Cambridge, UK
Tomáš Brázdil Masaryk University, Czech Republic
Markus Chimani Osnabrück University, Germany
Jan Černocký Brno University of Technology, Czech Republic
Eva Dokladalova ESIEE Paris, France
Jiří Filipovič Masaryk University, Czech Republic
Robert Ganian Vienna University of Technology, Austria
Dieter Gollmann TU Hamburg, Germany
Derek Groen University College London, UK
Juraj Hromkovič ETH Zürich, Switzerland
Ondej Jakl VŠB-TU Ostrava, Czech Republic
Hidde de Jong Inria, France
Zdeněk Kotásek Brno University of Technology, Czech Republic
Lukasz Kowalik University of Warsaw, Poland
VIII Organization
Steering Committee
Tomáš Vojnar, Chair Brno University of Technology, Czech Republic
Milan Češka Brno University of Technology, Czech Republic
Zdeněk Kotásek Brno University of Technology, Czech Republic
Mojmír Křetínský Masaryk University, Czech Republic
Antonín Kučera Masaryk University, Czech Republic
Luděk Matyska Masaryk University, Czech Republic
Organizing Committee
Radek Kočí, Chair Brno University of Technology, Czech Republic
Zdeněk Letko Brno University of Technology, Czech Republic
Jaroslav Rozman Brno University of Technology, Czech Republic
Hana Pluháčková Brno University of Technology, Czech Republic
Lenka Turoňová Brno University of Technology, Czech Republic
Additional Reviewers
Kfir Barhum Pavel Čeleda
Hans-Joachim Boeckenhauer Vojtěch Forejt
Yu-Fang Chen Lukáš Holík
Organization IX
Mapping Problems to Skills Combining Expert Opinion and Student Data . . . . 113
Juraj Nižnan, Radek Pelánek, and Jiří Řihák
1 Introduction
Building, running and maintaining enterprise networks is getting more compli-
cated and difficult. Part of the problem is the proliferation of real-time appli-
cations (i.e., voice, video, gaming) that demand more and more bandwidth and
low-latency connections. The growth of broadband networks, the continuously
increasing in link speed is driving end-users to an increasing interest towards such
applications. As a result, network management is getting everyday more chal-
lenging for network operators. In particular, voice over Internet Protocol (VoIP)
services have proved to be a disruptive technology that has transformed the
telecommunication industry. In this scenario, providing end-to-end quality rep-
resent a key point for VoIP service providers. Monitoring and improving network
c Springer International Publishing Switzerland 2014
P. Hliněný et al. (Eds.): MEMICS 2014, LNCS 8934, pp. 1–15, 2014.
DOI: 10.1007/978-3-319-14896-0 1
2 G. Antichi et al.
performance parameters like packet loss rate, delay or jitter is the widespread
approach to guarantee a good VoIP quality.
Such parameters can be monitored using either active or passive approaches.
In the first case, results may not represent the actual user experience as the net-
work performances may be corrupted by the test packets sent to estimate the
call quality. On the other hand, passive measurement tests observe user traffic
without increasing the network load. However, such an approach requires probes
able to process packets without loss. While this is a pretty easy task on low
speed links, it becomes more challenging on high capacity links (i.e. 1G, 10G
and more). This leads to a trade-off between the cost and performance of moni-
toring tools: those based upon dedicated hardware and those utilising lower-cost
but potentially lower performance COTS hardware.
In this scenario, we present an FPGA-based open-hardware approach to
enable QoS monitoring of VoIP services. We built our prototype on top of
the NetFPGA [17], an open networking hardware system that enables rapid
development of hardware-accelerated packet processing applications. The open-
source nature of the system allows an easy porting to different FPGA-based
platforms, making such a solution attractive regardless the current limitations
due to the physical hardware infrastructure used. The continuous rising of new
open-hardware platforms [10] [22] with a better hardware infrastructure increases
the potentiality of the proposed solution.
The paper presents the related work in section 2, while section 3 describes
the RTP detection scheme implemented for the proposed framework and reports
the results of its performance analysis. The strategies used for the estimation of
the IP level QoS parameters are summarised in section 4. Details of the architec-
ture of the developed open-source hardware platform are shown in section 5, while
section 6 presents the results of the performance analysis. Section 7 concludes the
paper.
2 Related Work
QoS monitoring of VoIP traffic is not a new field. Commercial tools, such as
[7], [6], [5] and [4] are availbale on the market. They are able to sustain high
traffic rates but on the other hand are closed-propretary and expesinve. There
are VoIP monitoring functions integrated in tools such as [8], but that is a
general tool for packet sniffing and monitoring, its purpose is not to monitor
VoIP continuously. Open source-software based solutions like [13] or [18] are
also available, However, in high speed scenarios (i.e., 1Gbps or higher) solutions
like that require expensive specialised hardware to work properly.
RTCP packets in the detection process can be a solution to the problem, since
RTCP packets have more predictable fields than RTP. Such an approach result
in poor detection performance when the RTCP packet rate is very low compared
to the RTP one. A third way is based on the recognition of statistical traffic pat-
terns (e.g., packet length, inter-packet arrival time). However, such a procedure
need a training phase or/and the knowledge of the statistical properties of RTP
flows to be effective. We propose a two-stage scheme, where the pre-filtering
stage is used to reduce the load of the second stage. Such an algorithm is able to
recognise an RTP flow without the support of any RTCP or signalling packet,
using heuristics that exploit the features of some RTP header fields.
In particular, as shown in Figure 1, the considered RTP detection scheme is
based on two different steps:
– Pre-Filtering : The Deterministic Filtering
– Flow analysis : The Heuristic Filtering
The first stage is used to deterministically pre-filter all the packets that
cannot belong to an RTP stream (e.g., TCP flows), while the second is actually
in charge of detecting RTP streams.
We define as a flow all packets belonging the same 6-tuple (IP destination
address, IP source address, layer 4 destination port, layer 4 source port, RTP
Payload Type, RTP Synchronisation Source Identifier). RTP Payload Type indi-
cates the format of the payload and determines its interpretation by the applica-
tion while RTP Synchronisation Source Identifier (SSRC) uniquely identifies the
source in a session and allow to distinguish different sessions of the same source.
It is worth highlighting that our detection schema separate each direction of a
flow because they can have different QoS parameters. Information taken from
each new arriving packet is compared with data acquired from previous packets
belonging the same flow. An increasing value of the supposed RTP header field
“Sequence Number” (SN) in consecutive packets of a same flow, could indicate
the flow might be RTP (at this stage we are not sure yet it is actually RTP). If
such a pattern is recognised in an adequate (we will define it better afterwards)
number of consecutive packets, the flow can be considered an RTP stream. There
are two main parameters that need to be carefully set in this process:
The need for a range of accepted SN values comes from the fact that packets
could be lost in the network or suffer from out-of-order arrivals. Accepting only
packets with perfectly consecutive SN values may lead to false negatives. NoP
represents the minimum number of received packets falling in the SNR necessary
to identify a flow as RTP avoiding False Positive (FP).
The pseudocode 1 shows how the heuristic filtering stage works. The counter
matching pkts is incremented every time the SN value of a new arriving packet
matches the proposed range. The stream is recognised as RTP when such a
counter reaches the N oP value.
An Open-Source Hardware Approach 5
PF P = (PIN F P )N oP (2)
F alse N egatives: In this scenario, the observed consecutive Sequence Num-
ber values are actually correlated, and in an ideal condition they are incremental.
However, it is possible to observe a non-incremental increase, even a decrease,
or an increment that is not in the acceptable range, when in presence of packet
6 G. Antichi et al.
loss or out-of-order arrivals. Such a phenomenon could lead to the result that a
packet, actually belonging to an RTP flow, does not pass the check proposed in
line 1 of algorithm 1. We define PAB as the probability that a packet belonging
to an RTP stream does not produce an increment of the matching pkts variable.
As a consequence, 1 − PAB is the probability to have the matching pkts (when
the packet really belongs to an RTP flow). Assuming each “packet in” event
independent, we can model the system as a sequence of independent Bernoulli
random variables. In this case, the probability PIN P R (i) of the event “to have
an increment of matching pkts at the i-th packet” in the case of RTP flow, can
be calculated using the geometric distribution:
i−1
PIN P R (i) = (1 − PAB )PAB f or i = 1, 2, · · · (3)
whereas the probability PIN P R of having an increment of matching pkts,
can be calculated as:
i=SN
R
i−1
PIN P R = (1 − PAB )PAB (4)
i=1
NF P
PF P = (5)
NOP
NF N
PF N = (6)
NOP
where
– NOP : the total number of RTP packets
– NF P : the number of non-RTP packets recognized as RTP
– NF N : the number of RTP packets not recognized as RTP
We used real traffic PCAP traces containing RTP streams to test the FN
probability. First, we analysed them with Wireshark [8] to obtain NOP for
each trace. Then, we compared the obtained results with the ones given by the
proposed scheme. The results of these tests showed that the proposed solution
detected all the RTP streams, and PF N was zero. We used then PCAP traffic
data that did not contain any RTP stream to test the FP probability obtaining
a FP probability of zero. In summary, all the performed tests report that the
100% of the RTP traffic has been recognised without any FN or FP.
Quality (PESQ) [3] is the most widely used intrusive measurement method to
evaluate the quality of VoIP applications. It is designed for LQ voice measure-
ment and the prediction of MOS value is based on a comparison between a
degraded speech signal and a reference one. Non-intrusive techniques do not
need reference data and can be used to monitor and/or predict the voice quality
directly from the network and system performance metrics (e.g., packet loss,
delay, jitter and codec). The ITU-T E-Model [1] is a computational model that
can be used to non-intrusively predict the voice quality. It allows the estimation
of the expected speech quality given some parameters as the codec, quantising
distortion or IP level QoS parameters [14]. The IP level QoS parameters mon-
itoring plays a key role in the control of the PQoS. In particular, most of the
techniques used for the monitoring of PQoS rely on the observation of:
– Jitter
– End-to-end delay
– Packet loss
4.1 Jitter
Jitter represents the variation in the inter-arrival time between packets. We
recursively calculate it using the following equations [21]:
|Di,i−1 | − Ji−1
Ji = Ji−1 + (7)
16
Di,i−1 = (Si − Si−1 ) − (Ri − Ri−1 ) (8)
where R is the packet arrival time and S the associated timestamp in the RTP
header. Notice that in order to obtain the jitter in seconds, we have to determine
the value in seconds of the timestamp unit. The RTP timestamp is generated
according to the RTP clock rate that is equal to the number of sampling periods
per seconds. As an example in the cases of G.711 or G.729 codec, the sample
clock is 8000 Hz, hence the timestamp unit is equal to 125 μs.
delayi − Ei−1
Ei = Ei−1 + (9)
16
delayi = rx timei − tx timei (10)
An Open-Source Hardware Approach 9
rx timei is the arrival time of the ith packet, while tx timei is the time when
the packet was sent. When a packet arrives, only the rx time can be estimated.
The tx time is obtained from the associated RTCP stream through the following
equations:
Where N T PRT CP is the NTP timestamp of the first RTCP packet received
(that is: the exact time it was sent, in seconds), and sRT CP is its RTP timestamp.
Thus, γi represents the time between the sending of the first RTCP packet and
the ith RTP packet (in seconds). This procedure requires a clock synchronisation
between the RTCP/RTP sources. The synchronisation can be obtained by means
of NTP servers when an accuracy of few ms can be tolerated in the estimation of
this parameter. Otherwise, advanced strategies for clock synchronisation must
be considered.
5 Architecture
We developed our architecture on top of the widely used, open-source, NetFPGA-
1G platform. For each RTP stream, statistics (i.e. jitter, packet loss and
end-to-end delay) are estimated in hardware allowing high precision traffic char-
acterisation.
free-running system clock, giving one increment every 8 ns (in contrast with the
96 ns inter-frame gap of Gigabit Ethernet). Unfortunately, this implementation
provides no means to correct oscillator frequency drift, bringing to a huge per-
formance degradation over time [9]. Such a problem can be solved by means of
Direct Digital Synthesis (DDS) [20], a technique by which arbitrarily variable
frequencies can be generated using FPGA-friendly, purely synchronous, digital
logic. Therefore, we decide to use the IRCT timestamp module proposed in [9].
Such a module proved to be both accurate and stable.
QoS Monitoring. The QoS Monitoring module detects RTP streams (section 3)
and gathers QoS statistics (section 4). Information related to each RTP stream is
stored in the SRAM memory. The NetFPGA SRAM memory is 4.5 MB, divided
in 512k words of 72 bit each. While half is reserved to the Output Queues, the
remaining is used by our data structures. Each stream requires 72B of memory
space to store flow data (IP addresses and ports pair), RTP/RTCP information
needed either for detection or statistic measurements as well as the calculated
statistics. The maximum amount of RTP streams managed at the same time by
the proposed solution comes from this memory limitation: using half SRAM we
can address approximately 32700 RTP flows at the same time. Moving the Out-
put Queues in DRAM or BRAM (the fastest on-chip memory usable in FPGA
design) would allows us to use the whole SRAM for our data structure, doubling
the maximum number of RTP managed at the same time. We point out that the
provided numbers are just board-specific and do not affect the way the overall
architecture works. Every time a packet of a new flow passes the deterministic
filtering, a new memory block in SRAM is needed. We perform a hash on the
flow data (IP source address and source port) to obtain the SRAM address to
store all the flow-related information. If a collision occurs (the SRAM address is
already in use) we check when the last packet of the recorded flow was received.
If the time passed is greater than τ we can suppose the recorded flow is finished,
and we can overwrite the SRAM memory block with the new flow. Otherwise,
we should perform a new hash function in order to find a new SRAM address
(we perform a maximum of four attempts, after which, the new flow will not
be recorded). The choice of τ parameter is crucial. Small values could lead to
the deletion of an active flow while big values could lead to memory leaks. We
decided to set it to 3 minutes [15]. Again, even in this case we decided not to rely
on the associated RTCP stream (i.e., the RTCP BYE packet indicates the end
of a conversation) as we want to assure the correct system operation without
the presence of RTCP packets.
6 Performance Evaluation
We evaluated the performance of the proposed solution in measuring the afore-
mentioned QoS parameters at different traffic loads. We used traffic traces sniffed
during real VoIP calls. In order to recreate as much as possible a real environ-
ment, we connected two VoIP terminals to Internet in different ways (i.e., wired,
12 G. Antichi et al.
wireless). We placed a traffic sniffer (i.e., Wireshark [8]) between the two termi-
nals in order to get useful traffic traces..
The obtained PCAP traces have been post-processed in order to estimate the
jitter and packet loss of RTP traffic produced in each direction by the considered
call. Then we replayed the PCAP traces into one of the NetFPGA ports using
an accurate hardware packet generator [12]. We point out the importance of
reproducing the traffic reported in the PCAP trace with high fidelity, in terms
of packets content and inter-departure times. Such a precision, guaranteed by
the NetFPGA packet generator, allows us to fairly evaluate the accuracy of the
statistics estimation of our system. Indeed, by means of this procedure we can
compare the statistics measured on-line by the NetFPGA with the results cal-
culated off-line using the PCAP file. These last results represent the reference
values. Figure 5 summarises the results for about 20 data sets; the circle repre-
sents the jitter value estimated on-line with the NetFPGA, whereas the x the
values calculated off-line on the PCAP file. The figure shows the high accuracy
of the NetFPGA system. Quantitatively, the average of the absolute error of the
NetFPGA solution with respect to the reference values is equal to 0.25%.
20
Our Prototype
Base line values
15
Jitter (ms)
10
0
0 5 10 15 20
Traces
As for the packet loss, the number of lost packets calculated by our sys-
tem was compared to that obtained with wireshark, obtaining exactly the same
values, with a 0% error.
An Open-Source Hardware Approach 13
Our Prototype
Netem values
500
400
End-to-End delay (ms)
300
200
100
0
0 5 10 15 20 25 30 35 40
Traces
We sinchronised the two clients and the NetFPGA system with the same
NTP server, and we started a VoIP call. We compared the on-line end-to-end
delay estimation performed by the NetFPGA with the values set in Netem during
each trial. The results summarised in figure (7) qualitatively show the accuracy
of the on-line estimation technique. From a quantitative perspective, the mean
absolute error is of 3.74%.
References
1. ITU-T G.107 - The E-Model, a computational model for use in transmission plan-
ning
2. ITU-T P.800.1 - Mean Opinion Score (MOS) terminology
3. ITU-T P.862 - Perceptual evaluation of speech quality (PESQ), an objective
method for end-to-end speech quality assessment of narrowband telephone net-
works and speech codecs
4. OmniPeek Network Analyzer. http://www.wildpackets.com/products/omnipeek
network analyzer
5. VoIP & Network Quality Manager. http://www.solarwinds.com/voip-network-
quality-manager.aspx
6. VoIP Monitor. http://www.voipmonitor.org
7. VoIPFuture. http://www.voipfuture.com
8. Wireshark protocol Analyzer (was Ethereal). http://www.wireshark.org
9. Antichi, G., Giordano, S., Miller, D.J., Moore, A.W.: Enabling open-source high
speed network monitoring on netfpga. In: Network Operations and Management
Symposium (NOMS). IEEE/IFIP (2012)
10. Blott, M., Ellithorpe, J., McKeown, N., Vissers, K., Zeng, H.: FPGA research
design platform fuels network advances. Xilinx Xcell Journal (73) (2010)
11. Costeux, J.L., Guyard, F., Bustos, A.M.: Detection and comparison of rtp
and skype traffic and performance. In: Global Telecommunications Conference
(GLOBECOM). IEEE (2006)
12. Covington, G.A., Gibb, G., Lockwood, J.W., Mckeown, N.: A packet generator on
the netfpga platform. In: Symposium on Field Programmable Custom Computing
Machines (FCCM). IEEE (2009)
13. Deri, L.: Open source voip traffic monitoring. In: System Administration and Net-
work Engineering Conference (SANE) (2006)
14. Garroppo, R.G., Giordano, S., Iacono, D., Tavanti, L.: Experimental and simu-
lation analysis of a wimax system in an emergency marine scenario. Computer
Communications (2011)
15. Guha, S., Daswani, N., Jain, R.: An experimental study of the skype peer-to-peer
voip system. In: International Workshop on Peer-to-Peer Systems (IPTPS) (2006)
16. Guntur, R.: Technique for identifying rtp based traffic in core routing switches. In:
Patent Application Number 20090135834 (2009)
An Open-Source Hardware Approach 15
17. Lockwood, J.W., McKeown, N., Watson, G., Gibb, G., Hartke, P., Naous, J.,
Raghuraman, R., Luo, J.: Netfpga-an open platform for gigabit-rate network
switching and routing. In: International Conference on Microelectronic Systems
Education (MSE). IEEE (2007)
18. Manousos, M., Apostolacos, S., Grammatikakis, I., Mexis, D., Kagklis, D.,
Sykas, E.D.: Voice-quality monitoring and control for voip. IEEE Internet Com-
puting, 35–42 (2005)
19. Hemminger, S.: Network emulation with netem. In: Australia’s 6th National Linux
Conference (2005)
20. Saul, P.: Direct digital synthesis. In: Circuits and Systems Tutorials (1996)
21. Schulzrinne, H., Casner, S., Frederick, R., Jacobson, V.: Rtp: A transport protocol
for real-time applications (2003)
22. Zilberman, N., Audzevich, Y., Covington, A., Moore, A.W.: NetFPGA SUME:
Toward Research Commodity 100Gb/s. IEEE Micro (2014)
Today’s Challenges for Embedded ASR
1 Introduction
Several years ago to deploy an embedded ASR application, the main challenge
was not ASR itself but the hardware, the ASR system was supposed to run
on. Affordable and fast mini PCs did not exist yet. Software-wise suitable HW
usually required expensive and time consuming porting and testing. Significantly
increased performance of affordable Linux based mini PCs allowed to design and
implement a simple, reliable, inexpensive, and especially today, widely usable
speech recognition systems for many different applications. Pocket-sized com-
puters can today easily outperform normal desktop PCs from some years ago
and unified development environments allows very fast porting to a new target
hardware.
Another limiting factor for many ASR applications was availability. ASR
systems were usually connected with some device like a PC, a car or installed
access point for ASR with microphone and button. For always available ASR
in a building a theoretical option was a set of microphones installed everywhere
but in a real life such a solution is not acceptable.
Changes in the mobile phones world several years ago helped to face this
problem. Most of the middle-class and state-of-the-art mobile devices today are
equipped also with wireless network support. With such a mobile device a user
does not need to access a microphone attached to a PC or some wall. Via wire-
less network the mobile phones can easily access also local ASR systems. So
c Springer International Publishing Switzerland 2014
P. Hliněný et al. (Eds.): MEMICS 2014, LNCS 8934, pp. 16–29, 2014.
DOI: 10.1007/978-3-319-14896-0 2
Today’s Challenges for Embedded ASR 17
mobile phones are acting as a remote microphone. Besides that, a change in user
behavior has occurred: The mobile phones are already accepted and being used
through all age groups [5].
Mobiles phones today are powerful enough to run even system for auto-
matic speech recognition. Unfortunately, the variety of mobile devices prevents
to design a low cost speech recognition software running on all available mobile
phones reliably. Indeed it is much easier to design a simple application which is
just recording the speech signal. Applying a client–server approach, the recorded
speech signal from the mobile device can be send to a recognition server for pro-
cessing. As already pointed out such a “server” can today also be a cheap device
which has a similar size as the mobile phone itself and can be installed wherever
ASR application is needed.
Today, speech recognition on a mobile phone or on a PC is a common feature.
Despite the fact that it is mainly a remote service the latency and the accuracy
is acceptable for given task. Why then use a local ASR system? We will answer
this question in Section 2. In Section 3 and Section 4 we show two examples of
local ASR systems with focus to aspects described in Section 2. In Section 5 a
brief summary is provided.
a valid sentence. Perhaps with very low confidence score but that’s not very
reliable rejection parameter. A low confidence score will even assigned to not
OOG utterance in a very noisy environment (car driving at a high speed,
people talking in the background, . . . ). Solution for OOG in the grammar
based system can be some simple garbage model as showed in Section 4 but
it makes the entire design even more complex.
2. Latency. There are 2 main sources of latency. The first one depends on vocab-
ulary size, the used technology and the service setting. It is latency generated
by the recognition system itself. More accurate or larger vocabulary usually
results in increased latency. For some applications, real time processing is
critical.
The second latency source is caused by communication with a remote service.
Such a latency can vary between 100ms and several seconds. In case we want
to use it while being mobile (e.,g. from within a moving car) it becomes
necessary to handle the case that the service is sometimes unavailable.
We tested the latency for one of the popular freely available ASR services for
commands covered by grammars used in Section 3 and Section 4. The best
latency was about 1.5 second, but 4 seconds latency was nothing special. As
we will show later, such a result is for the demonstrated real time applications
not acceptable. It is necessary to note that for the testing we used a good
Internet connection. On a mobile network in a moving car we expect even
worse numbers.
In the following two sections we show two real applications with a local
embedded ASR system and point out particular aspect of local grammar based
system in contrast to remote LM based ASR.
– light weighted
– simple and intuitive to use
– adaptable to physical and psychological changes
– offers various input methods like speech and touchscreens
– reliable
CONSTRUCTIVE TREATMENT
When a child shows he has not been given careful teaching relative
to sex hygiene, go to his mother and advise her to take the child to a
physician. Explain the physical as well as the moral and mental help
it may be to the child to have one of two very slight operations
performed, after which, with proper diet and bathing, the boy may
easily forget his wrong habits.
COMMENTS
Miss Vane saw a note fall upon Mary Pratt’s desk. She said,
“Mary, bring that note to me.”
The child, she knew had not yet read the note. Greatly
embarrassed, Mary looked questioningly at Clyde Mitchel before
starting toward Miss Vane.
Contrary to the courtesy which teachers Improper Notes
admonish pupils to show, Miss Vane stood
up, opened the note and perused it in the presence of the school.
While she was looking at the note, Clyde Mitchel buried his scarlet
face in his book.
“You wrote this note, didn’t you, Clyde?” asked Miss Vane.
Clyde only nodded “Yes,” and burrowed even deeper into his book.
“This is a shameful note,” said Miss Vane. “It contains words that
no child should ever write or speak. You may stay after school,
Clyde.”
The boys waited at the second corner from the school house for
Clyde after school.
In about ten minutes Clyde came running toward them.
“What did she do, Clyde?” they asked.
“Aw, nothing; she just preached a little and gave me a few licks
that wouldn’t hurt a baby.”
“What was in the note, anyway?”
He told them exactly what was in the note, and a loud “Hurray!”
went up from the group of listeners. The subject of conversation
among these boys as they went on down the street was as full of
unclean words and suggestions as the worst boys in the group could
think up.
CONSTRUCTIVE TREATMENT
If you can not deal with sex subjects privately, with pupils in the
lower grades, do not deal with them at all. Miss Vane made a mistake
in reading or referring to the note in the presence of others. In her
efforts to suppress such foul communications she occasioned a talk
upon the unnamable topics by all of her own room and many in other
rooms as well.
COMMENTS
CONSTRUCTIVE TREATMENT
Miss Terman should have drawn Pearl into the games of the other
girls early in the year. She should have said to the leader among the
girls, in private. “You have it in your hands to make a classmate
happy or miserable. You, yourself, will enjoy school better if no girl is
made sad and lonely. I know that the other girls will follow your lead
and, therefore, I desire that you invite Pearl Goodwin into your
school games and give her an opportunity to know and like good
company.”
COMMENTS
CONSTRUCTIVE TREATMENT
COMMENTS
Miss Fanson was a high school teacher who was justly admired by
the girls under her care. She had talked to the girls about the
deference and homage which they should show to their parents in
social matters. Alice Grant believed that Miss Fanson was exactly
right, hence was willing to act upon her teacher’s advice.
Since she had entered high school, boys had suddenly become very
interesting to Alice. She blushed one afternoon as she plucked up her
courage to reveal certain developments to her mother.
“Mother, the Freshmen are going to give a party, and a boy in my
German class has asked me to go. May I?” Her voice affected
indifference.
But Mrs. Grant knew her young daughter Retaining
and saw through that coolness. Her Alice Control
was excited and flushed and happy over a boy! And she stared
blankly for a moment as the realization forced its way. Then a
tempestuous refusal from a heart that resented her little girl’s
growing up sprang swiftly to her lips, but she kept back the words. It
did, indeed, hurt to have Alice begin to be a young lady, but could
even she, the most adoring of mothers, restrain time and the youth
that was blossoming in her child?
“I’ll have to think it over, Alice. I’ll tell you in the morning.”
And Alice went to her studying, confident that, whatever her
mother decided, she would be just and allow only big reasons to
weigh with her.
Mrs. Grant thought it over and that night talked it over with her
husband.
“She’s absurdly young—only fifteen,” he objected.
“Yes, but absurdly natural, too, and strong in her desires. I fear, if I
refuse, it may only surround boys with a mysterious glamour for her,
and she might then be tempted to associate with them in spite of me,
and any secrecy or deceit just now is dangerous. And you know our
Alice is growing pretty.”
Mr. Grant regretted and bemoaned the loss of his little girl, but
agreed. “But who is this boy?” he demanded. “Do you know him?”
“No. But I’m going to know all her friends from now on.”
And next morning, when Alice, pink-cheeked and eager-eyed,
sought her mother’s decision, she welcomed the “Yes” with a little
squeal of delight.
“But I’ve been thinking, Alice,” her mother added, “that I’d like to
know the boys and girls you’re going with. Wouldn’t you like to ask
some of them over here some evening before the party?”
“Would I? Well, rather! Mother, you’re a dear.”
“And what about a dress. I suppose you’d like a new one?” Further
question was stifled by an enthusiastic hug.
So they talked of the party and the dress, and then it was not far to
“the boys” and Alice’s new feeling for them. And Mrs. Grant felt that
the sweet intimacy she was entering with this new daughter more
than compensated for the loss of the little girl, who had suddenly
become a young woman.
When Alice returned from the party her mother showed interest in
each detail that her daughter related. She remarked: “You must have
had loads of fun—what did you have to eat? What did you especially
like in the conduct of your classmates?” It is while such concrete
subjects are being discussed that much guidance can be given the
daughter in her formation of opinions as to what is proper or
improper conduct. A teacher who brings about such intimacy as this
incident illustrates has done much for both mother and daughter.
ILLUSTRATION 2 (HIGH SCHOOL)
CONSTRUCTIVE TREATMENT
COMMENTS
The amative impulses of youth are not vicious, but need direction
and control. Self-control, above all else, is to be taught, and the
teaching must often be reinforced by wise, friendly restraint. Frank
friendships are to be encouraged; sickly, silly sentimentality laughed
out of court. If a teacher, instead of standing ready to give this help
and guidance when it is needed, encourages a sentimental devotion,
as Miss Kingsley did, the most fundamental safeguard of youth is
sacrificed—the ideal of controlled emotion, of a conscious saving of a
sacred experience for the future. A large range of interests, a healthy
balance of activities, and a wholesome unconsciousness of self, tend
to keep young people simple and child-like in their emotional lives.
Above all, no teacher has any business to give the impression that he
alone appreciates youth and its promise, or to make his relations
with impressionable boys and girls unduly personal.
Mr. Bradley was principal for two years of the Newcastle school.
He revealed his characteristics as a teacher so fully that we find in
him an example of the type not to be recommended and yet one that
is very instructive for students of school discipline.
In stature he was slightly below medium height. He came from
rural ancestry and was fairly well equipped as to physique. He had
black hair and eyes, somewhat mobile features and a wandering
gaze. His movements could hardly be called quick, but they were
prompt and without distinct mannerisms.
He had a most gracious manner when meeting people on the street
or in their homes. He spoke kindly to everyone and had the
reputation among the townspeople of being a royal, good fellow.
Even his pupils could not deny that he treated them very courteously
and jovially outside of school hours.
Despite all this he used essentially the method of the hen-pecking
incompetent when handling disciplinary matters in school. The
moment he entered the school precincts he was a different man. His
countenance then betrayed the sternness of the schoolmaster who
dwelt within and apart from the polite gentleman he seemed to be
when outside the school-room. His eyebrows gathered and his
muscles reverberated with the sense of authority that flooded his
whole nature.
His eye was on the lookout for misdemeanors and if a pupil made a
misstep in the realm where Mr. Bradley thought he had jurisdiction,
that harsh, strident voice, with but the slightest trace of fellow-
feeling, spoke the word of correction or announced an impending
penalty.
In the school-room it was his delight to slip up behind an offender
and pluck him by the ear as a reminder of duty. Being the only
instructor who indulged in this practice it soon came to be one of the
most odious signals of his presence in the room. When absorbed in
his subject he made instruction interesting; his pupils could not fail
to learn if they did not venture to vary the program by misconduct.
However, their recollection of his general attitude toward them, the
ease with which they could upset his plans by introducing a few
school pranks, the certainty that he would lose his temper on slight
provocation, always hung as a barrage screen between them and
undivided concentration on the subject-matter of their lessons.
Mr. Bradley made it a practice to watch for accumulating offenses.
He felt incompetent to handle minor evils, but attempted to squelch
a wayward pupil by reciting a list of grievances and applying
penalties for the same. He had a good memory for facts of this sort.
He could shake his finger in the face of a boy or girl and say, “Didn’t
you pull Esther’s hair yesterday ... trip up Jimmie on the way to class
in geometry and purposely spill the crayons when you were at the
board? Now, I have had enough of this. I want to know what you are
going to do about it.”
This gentleman could not catch the drift of things. Early in his first
year Mr. Bradley’s attention rested upon Ted. Ted was a short,
heavy-set chap of some fourteen years, incapable of any
revolutionary propensities, but able to interest himself with a variety
of aggravating tricks. His pranks were individually almost too small
to command severe penalties, but they were too annoying to escape
the principal’s eye.
Unfortunately, Mr. Bradley hit upon the lash as a cure for Ted.
Selecting a more pronounced misdemeanor as an opportunity for
settling accounts with the troublesome pupil, he gave him a sound
whipping.
There was some ground for the general protest that arose from the
high school. Ted was a favorite with every one. The crude principal
had struck one but he had wounded all. His untactfulness had made
him abhorrent to all, even to those who had not hitherto drawn upon
themselves his specific disapproval and useless punishments. Mr.
Bradley, perhaps, never knew that he had undermined his own
usefulness as much by this treatment of a school favorite as by any
single deed that transpired during his whole stay in Newcastle.
He had his own method of handling the problem of whispering. He
made it a rule that every pupil in high school must answer at roll call
at the end of the day on the matter of whispering. If a pupil had
whispered he must answer “Present,” and specify the number of
times during the day he had whispered. If he had a clear record on