Techniques in Speech Acoustics

327 Pages·1999·12.33 MB·English

by  Jonathan Harrington

Techniques in Speech Acoustics
Text, Speech and Language Technology
VOLUME 8
Series Editors
Nancy Ide, Vassar College, New York
Jean Veronis, Universite de Provence and CNRS, France
Editorial Board
Harald Baayen, Max Planck Institute for Psycholinguistics, The Netherlands
Kenneth W. Church, AT & T Bell Labs, New Jersey, USA
Judith Klavans, Columbia University, New York, USA
David T. Barnard, University of Regina, Canada
Dan Tufis, Romanian Academy of Sciences, Romania
Joaquim Llisterri, Universitat Autonoma de Barcelona, Spain
Stig Johansson, University of Oslo, Norway
Joseph Mariani, LIMSI-CNRS, France
The titles published in this series are listed at the end of this volume.
Techniques in
Speech Acoustics
by
Jonathan Harrington
Speech Hearing and Language Research Centre,
Macquarie University,
Sydney, Australia
and
Steve Cassidy
Speech Hearing and Language Research Centre,
Macquarie University,
Sydney, Australia
.....
"
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.
A c.I.P. Catalogue record for this book is available from the Library of Congress.
Additional material to this book can be downloaded [rom http://extras.springer.com.
ISBN 978-0-7923-5822-0 ISBN 978-94-011-4657-9 (eBook)
DOI 10.1007/978-94-011-4657-9
Published by Kluwer Academic Publishers,
P.O. Box 17,3300 AA Dordrecht, The Netherlands.
Sold and distributed in North, Central and South America
by Kluwer Academic Publishers,
101 Philip Drive, Norwell, MA 02061, U.S.A.
In ali other countries, sold and distributed
by Kluwer Academic Publishers,
P.O. Box 322, 3300 AH Dordrecht, The Netherlands.
Printed on acid-free paper
AII Rights Reserved
© 1999 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1999
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner.
For Alexander and Catriona
Owen and Nichola
CONTENTS
Preface ix
Vowel and Consonant Transcriptions xi
Contents of the CD-ROM xiii
1 The Scope of Speech Acoustics 1
1.1 What is speech acoustics? ........ . 1
1.2 The variable nature of speech . . . . . . . 2
1.3 Experimental designs in speech acoustics . 4
1.4 Some key areas of research in speech acoustics . 5
2 The Physics of Speech 9
2.1 Speech waveforms 9
2.2 Frequency analysis . 12
3 The Acoustic Theory of Speech Production 29
3.1 The source-filter decomposition of speech 30
3.2 The acoustic source in speech production 34
3.3 The acoustic filter in speech production 38
3.4 Vocal tract losses ..... 52
3.5 Radiated sound pressure. 53
3.6 The composite model. . . 55
4 Segmental and Prosodic Cues 57
4.1 Vowels ..... . 5i
4.2 Oral stops ... . i8
4.3 Nasal consonants 95
4.4 Fricatives . . . . 99
4.5 Approximants .. 105
4.6 Prosody and juncture 111
5 Time-Domain Analysis 131
5.1 Sampling and quantisation . 132
5.2 Definition of a digital signal 136
5.3 Simple operations on signals . 138
5.4 Windowing signals ..... . 140
5.5 Some common time-domain parameters 142
5.6 Convolution and time-domain filtering . 148
6 Frequency-Domain Analysis 157
6.1 Digital sinusoids ...... 15i
6.2 The discrete Fourier transform 162
6.3 Spectra derived from the DFT 164
6.4 Some points of procedure in applying a DFT 164
6.5 Spectral parameterisations . 1iO
6.6 Frequency-domain filtering. li8
7 Digital Formant Synthesis 195
i.1 Core structure of a formant synthesiser . 19i
i.2 Digital considerations 199
i.3 Periodic excitation .......... 199
7.4 Formant filter . . . . . . . . . . . . . 202
7.5 Combining the source with the filter 206
7.6 Parallel structure . . . . . . ..... 208
8 Linear Prediction of Speech 211
8.1 LPC and its relationship to digital speech 212
8.2 Techniques for calculating the LPC coefficients 215
8.3 Analysis of the error signal ....... 217
8.4 LPC-smoothed spectra and formants . . 219
8.5 Area functions and reflection coefficients 226
8.6 Speech synthesis from LPC-parameters . 233
9 Classification of Speech Data 239
9.1 Speech spaces and distance measures 240
9.2 Distributions of speech sounds 244
9.3 Discriminant functions and classification 250
9.4 Classification experiments 254
9.5 Classifying signals in time 261
9.6 Data reduction ...... 262
References 279
viii
PREFACE
This book is the development of a series of lectures to undergraduate and post
graduate students at Macquarie University on basic principles in acoustic pho
netics and speech signal processing. The first part of the book (Chapters 1 to
4) is intended to provide students with the ability to interpret acoustic records
of speech signals in their various forms. These chapters include a review of
elementary wave motion and frequency analysis as applied to speech (Chapter
2), a summary of the relationship between speech production and its acoustic
consequences (Chapter 3), and finally a review of the principal cues to speech
sounds and prosodic units (Chapter 4). The material from these first four chap
ters (and the related exercises on the accompanying CD-ROM) has formed the
basis of a one-semester undergraduate course in acoustic phonetics to students
primarily of linguistics, but also of other disciplines including computer science
and psychology.
The second part of the book provides an introduction to speech signal pro
cessing, which is intended for similar groups of students. It is therefore different
from more detailed introductory texts in this area (e.g., Rabiner & Schafer, 1978;
O'Shaughnessy, 1987) which assume both a background in engineering/signal
processing and a more sophisticated mathematical knowledge (e.g., Parsons,
1987). Part of the motivation for writing this section of the book is to make
many of the techniques and algorithms that are discussed in the engineering lit
erature on speech analysis (for example, in the IEEE Acoustics Speech and Signal
Processing journal) more accessible to both students and researchers of phonet
ics and speech science whose training is not usually in a scientific discipline. We
have therefore tried to keep equations to a minimum and to assume, as far as
possible. no more than a very basic understanding of algebra and trigonometry.
In this part of the book, we cover fundamental aspects of time and frequency
domain processing of speech signals (Chapters 5 and 6). Chapters 7 and 8 are
concerned with digital techniques for combining (in digital formant synthesis)
and separating (in linear predictive coding) the contributions of the source and
filter to the acoustic speech signal. The final chapter and related exercises deal
with techniques for the probabilistic classification of acoustic speech data that
forms the basis of more advanced work ill automatic speech recognition.
We have always felt that students of speech benefit from being able to carry
out their own experiments to test some of the theories that they learn about in
books such as these. However, preparing experiments that are also appropriate
to testing a particular aspect of speech acoustics can be very time-consuming to
construct and may not be feasible either because of the difficulties of collecting
(and digitising) large quantities of speech data or because of the problem of
student access to laboratory machines that are usually dedicated to research
projects. In the last three to four years, we have adapted a research tool that
has been developed in our laboratory (the mu+ system Harrington, Cassidy,
Fletcher, & McVeigh, 1993) for building and analysing speech corpora to the
needs of teaching on our undergraduate and postgraduate courses. In extending
this system, which is included together with a number of speech corpora on
a CD-ROM in this book, we have sought to develop an interface that would
obscure minimally the aim of the exercises, which is to solve problems in speech
acoustics rather than in computer programming. The revised software (known
as the EMU speech analysis system) runs on both UNIX and Windows 95 plat
forms. Since the EMU system can be used independently of the exercises in this
book for segmenting and labelling utterances, as well as analysing them in the
time and frequency domain (including spectrographic analysis), the exercises
can be easily modified to include other sets of speech corpora beyond those pro
vided on this CD-ROM. The EMU system also includes a set of extensions to
the Xlisp-Stat statistical system to facilitate the analysis of speech data; the ex
ercises on the CD-ROM use Xlisp-Stat extensively to allow students to analyse
real data from the various speech corpora included on the disc. The Xlisp-Stat
system (but not currently Emu) also runs on Macintosh systems.
We usually teach courses on speech acoustics after students have gained
some grounding in phonetics and phonology. Accordingly, we assume some
background knowledge of phonetic principles of sound classification and ele
mentary phonological theory, which are given excellent coverage in many other
books (see, for example, recent general phonetics texts at the end of Chapter 1)
and which are not dealt with in detail in this book.
x
Vowel and Consonant Transcriptions
The columns for the vowels apply to many talkers of American, Australian and
Southern British English (Adapted from Ladefoged, 1993).
Vowels
American Australian Southern British Key
English English English word
heed
hid
II Ia Ia here
u u u who'd
U U U hood
eI eI eI hay
£ £ r head
eJ ra ra there
a a a the
3' 3 3 heard
ou aU aU hoed
a '0 '0 hod
;) ;) ;) hawed
;)1 ;)1 ;)1 boy
A A A bud
re re re had
aI aI aI hide
ill a a hard
aU aU aU how
Consonants
Symbol Key Symbol Key Symbol Key
word word word
e
p pie think <t1 judge
t tie 0 these m my
k cat s so n no
b bat z zoo sing
D
d do J shoe w we
9 go 3 measure J run
f fan h he j you
y van tf chew 1 leaf
Xl