All images in this file are courtesy of MIT Press. Used with permission.
-I-
d-5z~I H5-T to.
A~Vac
s top
0.3
cDnsezea
W
/ Ag
il
CI
4;71
Y
E
-S
a0.2
I
,
,
for
I
r
0.I
0
-150
-100
-50
0
50
o00
TIME FROM CONSONANT RELEASE
(ms)
Figure 8.58 The various curves represent the assumed time variation of the cross-sectional
area of the glottal opening A, and the supraglottal constriction A, when a voiceless aspirated
stop consonant is produced in intervocalic position. In the case of the curves for A.s the solid
curve represents the trajectory As would have if there were no pressure increase in the mouth; it
is the same As that was assumed for /h/, in figure 8.34. The short-dashed curve represents the
actual dg that would occur as a consequence of the abducting forces due to the increased intraoral pressure for the stop consonant. The sloping straight line following the consonant release
(from = 0 to I = 70 ms) is an approximation to As(t) during this time interval and was used to
determine the calculated curves in figure 8.60. The supraglottal curves A(f) are intended to
approximate the changes in cross-sectional area for a velar consonant (dashed lines) and for a
labial consonant (solid lines).
+
Po
ONk
tc
Figure 8.59 Equivalent circuit used to calculate the flows and pressures at the release of a
voiceless aspirated stop consonant. Values of the elements are discussed in the text and in the
legend to figure 7.4. R1 = 1.5 acoustic ohms.
E
0
9L-
a::r
TIME RELATIVE TO CONSONANT
RELEASE (mns)
Figure 8.60 Calculated values of airflow versus time for intervocalic voiceless aspirated stop
consonants, corresponding to the time variation of glottal and supraglottal areas given in figure
8.58. Calculations are based on the equivalent circuit in figure 8.59, with a subglottal pressure of
8 cm H2 0. Element values for C, and R. are given in the legend for figure 7.4. The estimated
time of onset of vocal fold vibration is indicated by the arrow at 50 ms.
--- I-
-------------
�-----
t
/aL
-- - '-
/-
EWt
/a!'+
,/
ap
-kkf)
LIŽ1
5
0
I52X
1X
200
7­
TN-'t
_ s
_V 1
I
Y
_I_
ln
co
3hi8(
_
'-
L~4~L-J- 11
*0mge
aul
_ I L-L1-
=
I
I
SOO
1I
.
I
i
i
=
=
.­
-
-j-
I
4-
I
]~i
~/
C
v
.Z­
:=
i -1 I­
- - -Z
It
4L
- -
I
_
_
. I.
i
~
i~~~~~~n
o~ti---
.
I
Ii
2
4
I
II
/I
r AI--.
·
A
=
I
I
U1
ri:I
Pd1
i._ _
.
~TVs"
A
I
I -
.2~
II.
K)A
cc
fculte
4n
0C
Eaoc
3.
U
SC
s
presOJs
Ii.
I
E
U
LdJ
-J
V)
W
.­
z
TIME FROM CONSONANT RELEASE
(ms)
Figure 8.61 Calculations of airflow (top) and intraoral pressure (bottom) at the release of a
voiceless aspirated stop consonant. Curves are shown for airflow through the supraglottal con­
striction () and through the glottis (Us). Dashed lines represent estimates for velar release, and
solid lines for labial release. See text.
I--
-�-'T��-�-----l-�--�--�
�-�-�-`
, %
I
t
AtLe& S top coso0V\ W¥s
..
-
Figure 8.68 Low-frequency equivalent circuit for estimating flows and pressures in the vocal
tract for voiced obstruent consonants. The volume velocity source L4 accounts for the increase
in vocal tract volume due to active expansion of this volume. The resistance R is connected to
the circuit during fricatives and after release of the consonants.
Figure 8.69
Schematic representation of midsagittal section of the vocal tract at two points in
time during production of the voiced stop consonant Id/ in intervocalic position: at the time the
closure for the consonant is produced (solid contour), and immediately before the consonant is
released (dashed contour). The changing midsagittal contour is intended to illustrate the expan­
sion of the vocal tract volume during the closure interval for the consonant.
Z
1J­
VOICED
'OtI
VOICELESS STOP
STOP
5
5
/N
II
-00
00
.
OC
Y
n
.L
-100
0
PA
;. +
_
I
00
20
20
0
Z
10
Un
wE
I.
-
0
X,
-100
0
100
100
-100
0
-100
0
100
-100
0
100
-Inn
-Ivv
1.
V
- P
I0
0 _
<ME
_O_
0
Za
7Pm
1
,
r.
300
0
a
4
o/
/f
1
300
150
I
-100
I
0
I
ts.
100 .,
20
hi
U)
0
o00
_
MLU
a:
0
150
"3
i
w
- 00
_
S
I·
,
20
0
U.2
40
a hi
I
-20
a
,
,
I
-100
0
100
-20 °
|
f
Inn
VUU
TIME RELATIVE TO CONSONANTRELEASE(ms)
0-
Figure 8.70 Schematized representation of the time course of various articulatory and aero­
dynamic parameters during the production of an intervocalic voiced stop consonant (left) and
voiceless stop consonant (right), both consonants being unaspirated. The implosion of the con­
sonant occurs at time -100 ms on the abscissa and the consonant release occurs at time zero. An
intraoral pressure of 8 cm HzO is assumed. The top panel in each column gives the forward dis­
placement of the pharyngeal wall (data estimated from Perkell; 1969). The next panel shows the
estimated change in volume of the vocal tract during the dosure interval. The curves labeled
PASS. give the estimated passive volume change as a consequence of the increased intraoral
pressure; the curves P + A are the estimated combined volume increase due to active volume
change as well as passive change. The third panel gives the intraoral pressure P,. with the subglottal pressure indicated by the horizontal line. The fourth panel shows the estimated glottal
airflow. The bottom panel gives estimates of the percent change in vocal fold stiffness, relative
to the stiffness in the vowels, that occurs during the dosure interval and immediately following
the release.
-·
·--
-··
·- ··- ·
· II-
LI__---------�-----.---------'----
riTrmttive' tDAsos\aATS
Icm
I cm
Figure 8.1 Midsagittal sections obtained from cineradiographs during the production of the
three fricative consonants /f/, /s/, and As/, as indicated. The subject is an adult female speaker of
French. The approximate scale is given between the /f/ and /s/ configurations. In the case of the
/s/ configuration, an estimate of the cross-sectional shape at the constriction is given. (This shape
is drawn with an enlarged scale, as shown.) For each panel, two midsagittal sections for the
tongue (and, in the case of //, for the lips) are shown, representing the fricative in two different
phonetic environments. For /f/, the following vowel contexts are the high front rounded vowel
/y/ (solid line) and // (dashed line); for /s/, /a/ (solid line) and /i/ (dashed line); and for /Al,/a/
(solid line) and /u/ (dashed line). (After Bothorel et al, 1986.)
I)
E
-
cEAs
w
C)
0)
t?
(a
o
o
-200
-100
0
TIME RELATIVE TO RELEASE
100
(ms)
Figure 8.2 Estimates of the time course of the area of the glottal opening (A.) and the area of
the supraglottal constriction (A,) when a fricative consonant is produced in intervocalic position.
The solid lines indicate the area trajectories that would occur in the absence of a subglottal pres­
sure, and the dashed lines indicate the modified areas that are calculated when forces on the
structures due to the increased intraoral pressure are taken into account. The subglottal pressure
is assumed to be 8 cm H2 0.
IE
-J
TIME FROM RELEASED(ms)
Figure 8.3 Top: Time course of intraoral pressure (Pm) and transglottal pressure (AP,) corre­
sponding to the area trajectories for an intervocalic fricative consonant, in figure 8.2. The verti­
cal marks on the AP, curve indicate times when vocal fold vibration is expected to stop and to
start, that is, at about -105 ms and 5 ms. Bottom: Calculated airflow vs. time for the intervocalic
fricative consonant. Only the average flow is shown in the figure. There are fluctuations in the
flow when the vocal folds are vibrating. Offset and onset of vocal fold vibration are expected at
times indicated by the vertical marks.
-----------
,--·""-"-
_
.
/0- ~O­
I
:
300
4
.....
Ji InF 1
l]
I4t -isff
I
., -. 4
--
-
.
A
Ii
AVI \'
. P,
neluA>o
r "\'s
-UL.41.
L-:
� .Ab>.
VI III .I,-
1
I
I'
iI:
- _--l
'A,
I I
oo~~~l
!
/I
--
I
1..
II
V
V.1J
.
- P.r
/\
t \
- ----------------- ,-
-t,'A
--
Ae
wIVIP
y p
,
LU
-
I8
1l
I
ffl
I
N­
I -"
-I.
I - ---
­
9ffl
L­
/. ,
I
.
=44,.'
I
I
tAW" RRP~~~~~~~~~~~~~t~~~~'-~~~~~/j-----~~~~~~~~~~
I
.
I
M'
AA
fI10­
-,.4
I
"-��f-�
.
't'! 1W Ll
I
-
'#"!
in
.tJ
D
7-
6'
-d=1J
t
I
/N
A4
Ii-
I,'
·a X
I
E
7
I --
S
v1 ffri
I---j
A.-J. k -
I
L­
--
r
·
I'ii
l
I
!
IV
I-T Iv4Jv-
II
I i.-
:
Z~
i
II.-
m
m'"
..
I
V
·s)·'slC�gh�··
-
:
-44-
J - I I
I
I
. _ r.�,
1\J--
.
- .1
7,17"~dh~~b
Figure 8.12 A spectrogram of the utterance /afa/ is shown at the top. The spectra in the three
columns immediately below the spectrogram are sampled at several points in three regions of
the utterance: in the vowel immediately preceding the consonant (left, panels I to 4); within the
fricative region (middle, panels 5 to 8); and in the vowel immediately following the consonant
(right, panels 9 to 12). These spectra are obtained with a 6.4-ms Hamming window. In the vowel
regions, spectra are sampled at four adjacent glottal periods, whereas in the consonant the
spectra are sampled more sparsely. In panels 6 to 8, where there is no evidence of glottal vibra­
tion, the lower spectra are averages of a series of spectra obtained with a 6.4-ms window over a
20-ms interval. The upper curves in these panels are smoothed versions of the average spectra.
Waveforms and time windows, together with the times at which the spectra are sampled, are
shown below each spectrum panel.
·-
- --- - -···
· -·-- - ·
·
·-
I
Figure 8.16 Same as figure 8.12, except that the utterance is /osa/. Panels I to 4 represent
spectra (6.4-ms window) sampled at several times preceding the VC boundary, and panels 7 to
10 are sampled following the CV boundary. Panels 5 and 6 are average spectra calculated over
the first and second halves of the frication region, respectively. The spectra in panels A, B, C,
and D were obtained with a longer time window. See text.
-----------
·-
�-
I-----
---
-"---��--·.--I.--
-C-l^-·--"l
�.
-~~~~---------
-?­
LIPS
SOURCE P
I
I'
(a)
SUBUNGUAL
CAVITY
\
S
1
PALATAL
CHANNEL
BACK
CAVITY
.wor?4
tAo&Le
I c
I cm
(b)
cm
cm
the anterior por­
Figure 8.19 (a) Schematized representation of the detailed cavity shapes in
lower incisors,
the
at
located
be
to
tion of the vocal tract for //. The source of sound is assumed
from source
function
transfer
the
in
zeros
The
as indicated. (From Halle and Stevens, 1991b.) (b)
configuration
this
of
frequencies
natural
to mouth output in (a) are the
l
1-s i
-------------
I
�-----------`--I----------
Vbciedu&
QL Hi
lul
Is
V
If
WI
I
:1111In
i1
A
ricait\e-
II
Tffl#4A1
I ' -I l l
11
11hiff,
I
i
.
f
0
000
IC
wUl__
~~IWM
200
30
40 r3W500
"
fii
_Lll
I
1-1
-2
A
I I
1/1
o
.
_111
4
5n
_-Ah-
-'
'l
l
-¢
Ii
.2_
.
w
w,
I
I
.
.
-J
m
w
2,
W
>
W
-A
I­
100
200
300
I
I
i
L
!
500
I
v
200
100
200
. 1,
300
I_
- -.
C
I-
zl-
r
z
5
LI
-
~
~
/7-V-
TIME (ms)
I
L__
i.
.
Y
I.~ .
.
.
m-
Figure 8.76 Samples of acoustic data for the intervocalic voiced fricative consonant /z/ in the
utterance a zip. The spectrogram of the utterance is shown at the top left, and below this are: HI,
the amplitude of the first harmonic, measured as in figure 8.71; "noise," the amplitude of the frication noise, taken to be the peak spectrum amplitude (6.4-ms window) in the frequency range of
3.5 to 5 kHz; Fl, the first formant frequency, measured as in figure 8:71; the difference HI - H2
between the amplitudes (in decibels) of the first and second harmonics; and FO, the fundamental
frequency. The vertical lines are estimates of the points where intraoral pressure begins to build
up and where the pressure begins to decrease, based on examination of the acoustic record.
Spectra at selected points near the left and right boundaries of the fricative are displayed at the
right, together with waveforms and time windows (6.4-ms duration). The noise spectrum (panel
4) was obtained by averaging spectra over a 55-ms time interval All other spectra were calcu­
lated with a single time window.
__
_�I
~~~~~~~~~~~~~.-­
- ------- ·--------
CI--
,
A2 Al
A9
_, zzzLz
rnoD
Figure 8.25 Schematized model of the vocal tract for an affricate consonant. Closure is formed
by setting the area A1 to zero at the front of the constriction. The channel behind the con­
striction, with area A2, forms the fricative portion of the affricate after Al is released.
Figure 8.26 Conception of the nmidsagittal section of the anterior part of the vocal tract
prior to the initial release (dashed line) and 20 to 30 ms following the release (solid line) of a
palatoalveolar affricate.
tewz�n(e
Ynr
iL
I
,
'
A
g
E
E
E
0-
£
M
I
tI
L.&I
L,
1TI
T'Pr7
20
I
a0
0
TRE ()
TEW)
S1
V
I
I
-
.
ZIn - -
h
�--�`------��"��
I
1/1VC
, --Nrg
~-
~�
II
-
,~~~E,
pi>. /iI~
C~f
-"�...
... :....... ......
"a
�`....
i
`.. .�... ..._r
-
-
/h/
a-
&A-a-pirc-vn
b
uOiL
*
E
0.3
(J
0.2
(o0)
-j
_-
0.1
0w
I
III
B
0 100
10
0-.
(b)
E
50
aI
n
Hi
I
0
I
loo
200
300
TIME (ms)
Figure 8.34 (a) Estimate of time course of glottal cross-sectional area during the production of
/h/ in intervocalic position. (b) Calculated airflow corresponding to area variation in (a). A lung
pressure of 8 cm H2 0 is assumed. Resistance of subglottal airways is taken to be 1.5 acoustic
ohms.
Figure 8.33 Schematization of glottal opening for an abducted glottal configuration. The
glottal width at the level of the arytenoid cartilages is w,; ,m and {c are the lengths of the mem­
branous and cartilaginous portions of the glottis.
-
600
MODAL
­
400
(a)
>. 200
I--
o
,2
O
I
4
8
12
16
20
2'4
I
b)
0
TIME (s)
0
FREQUENCY
500
o0O
(Hz)
Figure 8.35 (a) and (b) Schematic representation of the waveforms and (c) and (d) spectra for
modal voicing and for breathy voicing. For breathy-toicing, the amplitude of the fundamental
component is enhanced relative to the amplitudes of higher-frequency components. A funda­
mental frequency of 125 Hz is assumed. The amplitudes of the waveforms and of the spectral
components are selected to be in the range observed for male voices.
--------~------�
·-··-
----7--·
----·----------------
--
-------­
I
-e-
so
I
2
1
- -
-
0
3
2
1
I·
.
. . ......
,-.
/~
~
J
L
~-An.
AL
.
- -
_--.
I1
4
5
FREOani
. .__
.
--
- - -'i
I
3
2RE1
f---
<o
1
I
-
'Y
,.E
'
A
IlA =-m bn
ICC�
RA
---
A
J
om
Jo
I-
WV
T*Cw-')lu·
,-
W'
- -
IiWV
AL
W
9
970 ma
ms
401
J.I
IA
,>
1..
-A,~~
&?=
A. LR 1 _
__v.-,-
ax
�mm
890
"NI
�A..t/U'
. �_
- , 4-_i-'-.
V-5"
'q
nr---
r· I
r1000
A,,..
U""
�
I
I
1000m
a
Figure 8.36 Spectra sampled near the onset of the vowel (upper panels) and about 30 ms later
(lower panels) in the syllable /ho/ produced by two speakers: a female (left) and a male (right).
Waveforms are shown below each spectrum. together with the time window over which the
discrete Fourier transform is calculated.
I
I
I
I
to pdrinc
60 ?o\
*,so
I-
o
source
xyJ noise $
IE
n
Don D&ilc
periodic
af glot, s
o
40 L-
) Lk ?L(
.l)0
SOURCES OF
Z
-
_Jo
,0 n X
30
V
o
noise source
o o
I
r
I
I
1
2
3
4
FREQUENCY
(kHz)
Figure 8.41 Calculated spectra of radiated sound (before adding the effect of an all-pole trans­
fer function) due to sources of turbulence noise in the laryngeal region during /h/ and to the
periodic glottal source at a time when the glottis is in a configuration for modal voicing in an
adjacent vowel. For the periodic source, the solid line gives the levels in 300-Hz bands, whereas
the open circles give the levels of every fourth harmonic, assuming a fundamental frequency of
125 Hz. The spectrum of the noise is in 300-Hz bands. The distance from the mouth is 20 cm.
To obtain the spectra for a vowel, the all-pole component of the vocal tract transfer function
must be added to these spectra. See text.
GLOTI
Icm
Figure 8.38 Schematized views of the laryngeal region in the sagittal plane (left) and in coro­
nal section (right), indicating how airflow through the glottis nmight impinge on the surface of
the epiglottis (left) or on the ventricular folds (right) to produce turbulence noise that is repre­
sented as a source of sound pressure.
-
-·-
-----------�
�-
12 OC
4
0
C
o
_
1
2
_m
-
3
FREQ (kHz)
4
5
4
5i
Ci
4
(a)
o
1
a
r
h
i
2
3
FREOQ
(z)
AODE
MODEL
Z/-
U
Figure 8.44 Examples of smoothed spectra sampled in /h/ (solid lines) and in the follow
ing vowel (dotted lines) for the utterances /he/ (male speaker) and /ho/ (female speaker) a
indicated. Spectra were obtained by calculating a discrete Fourier transform (time window c
-30 ms) and then smoothing this spectrum with a weighted frequency window of width abou
400 Hz.
A
(b)
>
Figure 8.50 (a) Midsagittal section of vocal trad for vowel Ai/, showing constriction in the
palatal region. (From Perkell 1969.) (b) Schematization of the airway with two constrictionsone at the glottis, with cross-sectional area A, and the other in the supraglottal airway, with
cross-sectional area A.
--
[h
[60
o
0o
2
FREO
3
z)
4
5
4
5
-[h[h
0
o
[h]
'C
-
0
1
2
3
FREQO(kHz)
IFRU (kHz)
Figure 8.54 Some smoothed spectra sampled in the syllables /hi/ and /hu/. For each utterance,
two spectra are shown: a spectrum sampled within /h/ (solid line) and a spectrum sampled in the
following vowel (dotted line). The prominence of the FZ or F3 peak in the noise spectrum is
indicated in each panel. The upper row represents a male speaker, and the lower row a female
speaker.
1
-------
�
�
--------
-
-------------�I
FREQ Oz)
1
-
It
I
o
S
I
20
I
/,,,
^
.riL
I5
40
60
1AAAh
I
80
100
TIME (mns)
E
FREO (kHz)
,IVAW vVV
47 '
'
Y,
aam
'
0
20
40
I
20
40TIME (ms)
TIME (s)
60
80
o1
60
80
o100
Figure 8.57 Displays of several aspects of the acoustic signal in the vicinity of voicing onset
in the syllable /he/ produced by a female speaker (top) and a male speaker (bottom). The upper
panel of each section gives the spectrum (30-ms window) sampled near the onset of glottal
vibration (left) and 30 to 40 ms later (right), as indicated by the waveforms immediately below
the spectra Below these waveforms is the complete waveform for about 100 ms near voicing
onset. The bottom waveform in each section was obtained by passing the signal through a
bandpass filter (bandwidth = 600 Hz) centered on the third formant. This waveform is less regu­
lar near voicing onset than it is later, indicating the dominance of the noise component in the
glottal source.
^--I------
------"c�-
-�