Dept. for Speech, Music and Hearing

Quarterly Progress andStatus Report

Vowel duration and a modelof lip mandible coordination

Lindblom, B.

journal: STL-QPSRvolume: 8number: 4year: 1967pages: 001-029

http://www.speech.kth.se/qpsr

I. SPEECH PRODUCTION

A. V O W E L DURATION AND A MODEL O F LIP MANDIBLE COORDINATION

B. Lindblom

A c lass ica l problem in phonetics is that of vowel duration. The

problem a r i s e s because of the systematic variations that the acoustic

duration o i a vowel is known to exhibit a s a function of the feature

composition of both the vowel itself and the adjacent consonantal en-

vironment. F o r example, everything e lse being equal, I (1) open vowels tend to be longer than close vowels;

( 2 ) a vowel is generally longer before a voiced consonant than before a voiceless consonant; and !

(3) it is longer also before a fricative than before a stop (1, 2) I I

At present the causes underlying the differences listed under i !

(1) - (3) a r e largely unknown. Nor can anything be said with cer tain-

ty about the i r cross-lingual generality. Consequently phonetic theory

has yet to be developed to the point where i t predicts and explains the

acoustic facts of vowel duration and provides a rationale for regard-

ing them a s linguistically determined o r a s consequences of universal

physiological conditions on human speech production. In so f a r a s

.. such an undertaking is successful i t i s likely to deepen our general

understanding of the motor organization of the syllable and to b e a r on

the question of determining the fo rm of phonetic rules in linguistic de-

s c ription. I I

In the following study we attempt to analyze open and close vowels

in t e r m s of the i r production. It is shown that a dynamic model of l ip

and mandible interaction can be constructed that predicts the observed

durational difference. The assumptions on which these predictions

have to be based a r e then exposed to data on l ip and mandible move-

ment.

Open-close dimension of vowels

Traditionally the degree of opening of a vowel is said to be de ter -

mined by i t s "tongue height". It has been recognized, however, that

the depression of the mandible can a lso se rve a s a c r i te r ion of open-

ness . But since i t i s easy to demonstrate that perceptually acceptable

STL-QPSR 4/1947 2.

vowels can be produced a lso with a pencil between the teeth, tongue

height ra ther than jaw opening has been considered the pr imary

feature controlled by the talker.

Nevertheless, a fairly good agreement between classifications

based on tongue height and jaw opening i s demonstrated by the resul ts

of the following experiment: Three subjects were asked to sustain

Swedish vowels (long and short). A photographic method was used to

record the position of the mandible which was indicated by a special ( 3 1 device . I

Fig. I-A-1 shows the depression of the mandible a s a function of

the th ree natural c l a s ses of front, unrounded and front, rounded I

and back, rounded vowels, Each point represents the mean of ap-

proximately 20 measurements f r o m the th ree subjects. Lines have

been drawn to connect vowels assigned to three c l a s ses of opening.

These plots b e a r a ra ther c lose resemblance to c lass ica l vowel quad-

r i la terals . j

Vowel duration and mandible ~ o s i t i o n

Since the inherent duration of a vowel has been associated with its 1 tongue height, and tongue height and jaw opening appear to be co r re -

lated in non-compensatory modes of pronunciation, it i s natural t o

expect a t least an approximate correlat ion also between vowel dura-

tion and mandible position. Two ta lkers who participated in the 1 above-mentioned experiment a l so read l i s t s consisting of randomized

sequences of nonsense words containing long and short Swedish vow-

e ls in an [I ' b-b(b)1] f rame. F o r each ta lker , the duration of the

vowel segment is defined a s the interval between the plosion of the

f i r s t [b] and the initiation of the occlusion for the second [b]. The

resul ts of these measurements (pooled and averaged) a r e presented

in Fig. I-A-2 and compared with data f r o m the f i r s t experiment. It

is c l ea r that the expected correlation, although g ross , is present i n

these data.

A dynamic model of labial articulation

An ar t iculatory interpretation of the vowel duration measurements

of Fig. I-A-2 where the vowels occurred i n the context of [I' b-b(b)I]

can be made most easi ly in t e r m s of the midsagittal separation of

GRAVE: FLAT:

LONG VOWELS

GRAVE: FLAT:

SHORT VOWELS

Fig. I-A. 1 . The depression of the mandible as a function of the three natural c lasses of front, unrounded [-grave, -flat] and front, rounded [-grave, +flat] and back, rounded [tgrave, +flat] vowels. Each symbol i s based on a pooled average of at least 20 measurements from three talkers.

DEGREE OF JAW OPENING

Fig. I-A-2. Relation between vowel duration in the context of [I'b-b(b)1] and degree of jaw opening for two talkers. The vowel duration data a r e pooled averages of 20 measurements for each vowel. The jaw data a r e based on approximately 15 measurements and a r e f rom the experiment associated with Fig. I-A- 1.

STL-QPSR 4/1967 3.

the lips: When this parameter is ze ro we have a bilabial occlusion;

when the parameter a s sumes a positive value we have an open labial

configuration a s during a vowel. Consider a point located midsagit-

tally on the upper l ip near say the vermilion border and a correspond-

ing point on the lower lip. In the following we shall observe l ip move-

ments in t e r m s of the displacements of these points. In relation to

some reference, for example the maxilla, the displacement of the

upper lip is due mainly to labiomuscular forces. The lower lip, on

the other hand, moves owing to the presence of labial and/or mandi-

bular muscular forces .

Assume that the upper and lower l ips and the mandible can each ( 6 ) be represented by a damped spring-mass sys tem . F o r such a I

sys tem the displacement x(t) i s governed by the standard cquati-n

of motion:

+ b& t kx = f(t) ( 1)

where m , b, and k a r e the m a s s , friction and spring constants and

f(t) the driving force. Dots above x represent differentiation with

respect to the t ime variable t . Eq. (1) could for instance be applied

to the upper lip. The lower l ip and the mandible on the other hand I a r e coupled and thus the differential equations describing their mo-

! tion differ frorn Eq. (1). The interaction between the two bodies be- ! comes negligible under cer ta in physical conditions, however. Thus, 1

i I if the m a s s of the mandibular s t ruc tures i s assumed to be la rge , and

i t s resonance frequency smal l in comparison with the corresponding

attributes of the lower lip, the motion of the mandible will i n a f i r s t

approximation be unaffected by the l ip and the movement of the lower

l ip will be the s u m of the individual l ip and jaw components. Since

i t appears justified to suppose that in a f i r s t approximation, the l ips

and the mandible come reasonably close to meeting these assump-

tions under physiological conditions we define s( t ) , the midsagittal

separation of the l ips, a s

~ ( t ) = xu(t) - ~ ~ ( t ) - xj(t) - x 0 ( 2 )

This equation i s by definition valid only when

xu(t) - xR(t) - xj(t) x0 > 0. F o r x (t) - xQ(t) - xj(t) - xo < 0 we se t u

s(t) = 0. The t e r m s xu and x represent the labio-muscular displace- R ments of the upper and lower l ips and x . denotes the jaw-dependent

J

depression of the lower l ip relative to the neutral location of the upper

l ip and t is t ime. In accordance with the preceding discussion x and a x . a r e he re t reated a s independent additive t e rms . These pa ramete r s

J a r e i l lustrated i n Fig. I-A-3. Elevation above the neutral o r jaw-

dependent position makes a parameter positive and depression makes !

i t negative. Before inser t ion into Eq. (2) the signs of xu, xi and x j

I

i n Fig. I-A-3 a r e thus negative, positive and negative, respectively.

The t e r m xo is a constant which stands f o r the uncompressed length I

of the t i ssue between the points on the l ips a t the moment when s be- I I

comes ze ro o r the lips just touch. Evidently x - x. > x when U - ~ R J 0

I I

the mouth i s open, and for closure: x 5 x These situa- u - X~ - Xj 0'

tions a r e depicted i n the two halves of Fig. I-A-4. The fact that we 21-

~ G W x - x . o r the distance between the points on the lips to be- U - ~ R J

come sma l l e r than x presupposes a cer ta in compressibil i ty of the i r 0

marginal t issue. In the right half of Fig. I-A-4 this c ircumstance is

taken into account by the presence of the spring connecting the two

bodies. In the argument to follow below this element will be assumed

to be ze ro but have the length of xo.

Simulation of vowel duration

The model developed above and il lustrated in Figs. I-A-3 and

I-A-4 can now be used to compute the hypothetical course of the mid-

sagittal separation of the "lips", s(t) , for say, [ b ~ b ] syllables to

which we assign various degrees of vowel o r jaw opening. In these

computations we shal l rely on a repeated application of Eq. (1) which

will enable us to der ive the component motions of s(t) , viz. , xu(t), - xA(t) and x.(t). F r o m Eq. (2) we shal l then obtain the t ime functions

J representing s(t) which can be studied with respect to "acoustic vowel"

duration a s a function of "jaw" opening.

Our f i r s t s tep is to assign cer ta in "labial" and "mandibular" target

positions to each segment of [bVb]. F o r [b] the upper "lip" is de-

pressed to make contact with the lower "lip". F o r the vowel i t is

raised. The lower "lip" is elevated fo r [b] and lowered for the vowel.

The mandible i s ra ised for [b) and assumes th ree degrees of opening

fo r the V which is to be simulated a s a "close", "half-open", and

I I open" vowel. The notion of articulatory target position is implicit

in traditional phonetics which associates with each phoneme in a given

LABIO- MUSCULAR x " / ~ - - - - - l

LABIO- v, - - l x j JAW- M U S C U L A R n DEPENDENT

Fig. I-A-3. Graphical il lustration of labial and mandibular parameters of s , the midsagittal separation of the lips. x and xL represent the labio-musculai displacement of the upper and lower lips yelative to the maxilla and the mandible respectively. x. denotes the jaw-dependent displacement of the lower lip relative to the neutral ldcation of the upper lip. The combined distances of x and x f rom the l ip m a r - L gins a r e represented by xo. The a r rows indicate the dyrection of the displace- ment. Upward displacements occur in the positive x-direction. F o r the particul- a r labial configuration schematized i n this figure the values of x and x . a r e thus negative whereas x is positive. a

J

Fig. I-A-4. The parameters and configuration of Fig. I-A-3 interpreted i n t e r m s of mechanical models of the lips. Upward displacements occur i n the positive x-direction. x i s the displacement f r o m a neutral position that resul ts f rom applying a force t o thUe upper "lip". x. i s the "jaw1'-dependent position of the lower "lip" and x i s the dis- placement a w a j f r o m this position that i s due to the application of a "ladial" force. In the left pa r t the ayetame a r e independent; i n the right half they interact. The constant x is equal t o the uncompreaaed length of the spring connecting the two bodiee in t%e right par t of the figure.

STL-QPSR 4/1967

language cer ta in ideal positions of the articulatory organs and inter-

p re t s syllables, words, phrases etc. basically a s sequences of I move- I

ments between such positions (cf. ~ i e v a s ' s "Stellungslaute und Uber- 8

1

gangslaute" (8)). F o r a given ar t iculatory component to reach i t s I

t a rge t positions i n a sequence such a s [ b ~ b ] a t ime-varying force

must be applied to it. F o r simplicity we make the value of this force

constant for the duration of each phoneme. As a resul t the syllabic I I

I

pattern of the force becomes rectangular i n shape. The rectangular

pulses shown i n the left par t of Fig. I-A-5 represent such stylized

fo rces applied to the upper and lower "lips" and the "mandible" i n

[ bVb] . In t e r m s of these patterns "vowel duration" can be defined

"articulatorily" a s the interval during which the force assumes the

value appropriate f o r ther~vowel'.' In Fig. I-A-5 this interval is de-

noted DA fo r the "lips" and D. f o r the "mandible". The response of J

the "lips", o r the "mandible", to any given input force, f(t), can be

determined f r o m Eq. (1) fo r instance by means of Laplace t r ans -

forms (7, 9 ) once the values of the constants m, b, and k have been -----7

chosen. F o r computational convenience b was se t equal to 2 Jkm

i n deriving the displacements of these systems. This choice makes -at

them cri t ical ly damped and gives them an impulse response of t e

w h e r e o = Jk/m. The general f o r m of the response of a crit ically

damped and uncoupled second-order sys tem to a rectangular force

pulse is

-o(t-a)-e-o(t-a) ~ ( t ) = A. + h ( t - a ) [ l - a(t-a)e 3 -

where A i s the initial location of the system; A the total extent of 0

displacement; o determines the ra te of movement; t is t ime and - - u(t-a) and u(t-b) represent the conventional notation of unit stept that

occur at t=a and t=b respectively. By definition these functions a s -

sume the values of z e r o for ( t -a) < 0 and (t-b) < 0 and one for

(t-a) 2 0 and (t-b) 0. In the present application t=a and t=b stanci

f : ~ r the moments of onset of "phoneme commands". Eq. (3 ) has been

tested empirically and found quite real is t ic ( lo) . (AS an alternative

to Eq. ( 3 ) the response of an overdamped second-order sys tem might

a l so be considered. ) The curves in the right half of Fig. I-A-4 a l l

have the f o r m specified by Eq. ( 3 ) and represent responses to the

UPPER LIP

I I

I

~ D I + LOWER LIP I I I I I

I I MANDIBLE

--Dj+

J

time-

UPPER LIP

JAW- DEPENDENT I

x j (t) I I

1 1 I I I 1 I I I I

3 I I I

l e D i 7 time-w

<

Fig. I-A-5. Time-variations of input forces to the model of Fig. I-A-4 for a hypo- thetical [bVb] syllable (left part). The model components a r e taken to be second-order and non-oscillatory systems. The upper lip sys tem responds to the input force a s shown in the right half of the figure. The displacement of the lower l ip contains the labio-muscular and a jaw- dependent components indicated below to the right. Three degrees of vowel opening a r e reflected i n the mandibular and jaw-dependent curves.

I

To summarize, a dynamic model of labial and mandibular coordi-

nation can be constructed in such a way that i t correct ly predicts a

durational difference between open and close vowels in [ b ~ b ] type of I

syllables. This resul t is based on a number of simplifying assump- 1 tions concerning the shape of the input forces to the l ip and jaw sys-

tems and the dynamic propert ies of these systems. It is c l ea r that

fur ther constraints must be imposed on the choice of t ime constants,

impulse response etc. before the empirically observed correlat ion 1

between vowel duration and mandible position can be predicted with

quantitative precision. The major purpose of Fig, I-A-5 and Fig.

I-A-6 and the associated discussion, however, has been to demon-

s t r a t e that the durational variance of open and close vowels is not i necessari ly a character is t ic of the input control signals to the model.

Neither D . nor D were varied a s a function of vowel opening. It can J R

be generated simply a s a resul t of the superposition of mandibular

movement on the opening and closing gestures of the lips. These

findings a r e in qualitative agreement with the data of Fig. I-A-2 and

suggest the hypothesis that the variability of the duration of open and

close vowels should be attributed to mandibular coarticulation. It

now remains to be seen to what extent this hypothesis is compatible

a lso with data on lip and mandible movement.

Experimental method and procedures I A procedure has been developed to record lip and jaw movements

continuously and in synchrony with the speech signal. This procedure

which is an improved version of the method described ea r l i e r by I

I

Lindblom and Bivner ( I4) involves the use of cylindrical miniature

lamps which a r e about 0.5 m m in diameter and 3.0 m m long, and

a r e attached to the t a l k e r ' s l ips and to a special device indicating

the position of the mandible. Similar techniques have been described

ea r l i e r by Jeaff resf)r, ( I5 ) and more recently by Kozhevnikov and

Chistovich ( I6) . The movement of the lamps during speech i s r e -

gis tered by a 35 mm oscilloscope camera in which the f i lm runs

perpendicular to the movement dimension under analysis.

The camera was run at a speed of approximately 2 cm/sec.

Since the noise of the camera was considerable i t was placed behind

a sound-insulating door in a room adjacent to that used by the subject

HALF-OPEN

time

Fig. I-A-6. The course of s(t) , the midsagittal separat ion of the lips in [ b ~ b ] syllable simulated for th ree degrees of vowel opening. Acoustic vow- el duration, D.v, can be defined a s the interval during which s > 0. This interval i s seen t o increase a s a function of vowel opening.

STL-OPSR d l 9 6 7 8 .

and the experimenters. The light from the lamps reached the camera

by way of a circular window in the door. Thanks to this arrangement

a signal-to-noise ratio better than 40 dB was measured on the tape

recordings. Spectrograms of speech recorded during the experi-

mental sessions showed no t race of interfering noise, Fig, I-A-7.

To minimize the superposition of head and body movements on the

film records, a head clamp device was attached to the sound-insulat-

ing door. This device consisted of a curved metal bar. In the pres-

ent experiments it was oriented so that the subject faced the camera

when he pressed his forehead against it. Its distance to the camera

could be adjusted along two f i rm horizontal metal rods. These rods

were parallel to the main axis of the camera lens system and could

thus serve a s references when a subject was positioned in relation

to the camera. Anatomical reference lines such a s the Frankfurt

horizontal (17, 18) were marked on the subject's head to facilitate a

suitable and reproducible relative positioning. To fix the subject' s

head against the head rest , leather straps with an adjustable buckle

were also used.

In the experiments to be reported below measurements were made

of the vertical movements of the lips and the jaw. Lamps were at-

tached midsagittally to the lips and a device for indicating the man-

dible. Since slight head and body movements could not be entirely

eliminated in spite of the above-mentioned measures i t proved desir-

able to introduce a reference lamp that would follow such movements.

This lamp was located just above the upper lip in approximately the

same coronal plane a s the lips and the mandible lamps. This was a r -

ranged with the aid of a pair of tight-fitting glasses provided with a

trunk-like extension for holding the lamp. The lip lamps were at-

tached by means of a drop of glue only insignificantly larger than

the contact a rea and in such a way that they would always be separated

even during a [b] -closure. The cables leading to the lamps were thin

and extra flexible. It i s clear that the skin of the chin can movc con-

siderably in relation to the mandible a s the muscles of the lips con-

tract. Consequently it cannot be used a s an accurate indication of

mandible position. For this purpose dental casts were made for

each subject individually and from these, cap splints. The cap

splints were thin and light and fitted tightly to the lower incisors,

kHz -

1- -

n,,

0- i I ( I ' I ' I ' I ' .-2 11 0 3 2 .3 .1 .5 .6 sec.

Fig. I - A - 7 . Spec t rogram of [ I ' ba:b1]. At top the synchronization pulses .

STL-QPSR 4/1967 9.

A piece of f i r m wire was attached midsagittally and ventrally to each

such device. It was adjusted s o that the l ips would touch it only i n I

bilabial stops fo r which i t s ver t ical location would coincide roughly 1 with the level of l ip contact. I

Before the sessions the subjects had plenty of t ime to get used to I

talking with the lamps and the jaw device on. After they had had

some shor t pract ice it was not possible to hear whether they were

talking with o r without lamps and cap splint. Listening to tape-re-

c o r d i n g ~ of the t e s t lists confirms that impression. Moreover, f i lm

records of words f r o m these l i s t s pronounced with and without the 1 cap splint fail to reveal radical differences between the two conditions,

(Fig. I-A-8). The vert ical distance between the lamps was measured

fo r th ree ta lkers and fo r both conditions i n sustained versions of [i:]

and [ a :] which occur in the present speech mater ials . The differ-

ences fail to reach significance a t the 5 70 level. In Fig. I-A-8 a I comparison is made between normalized and averaged records of 1 the ver t ical distance between the l ip lamps to a s s e s s the effect of the

cap splint on the movement of the lamps i n [I' bi:bI] . As can be seen

the curves f o r the two conditions a r e almost identical. The normali-

zation and averaging procedure will be described below i n the sec-

tion on Experimental Results. 1

F o r synchronization purposes the cur rent through the lamps was

pulsed. The speech signal would thus be recorded on one of the

channels of a twin-track tape- recorder the other channel being used

for the pulses. The lamps were l i t by the experimenter for each i t em

i n the t e s t l is t individually, To facilitate the detailed temporal align-

ment of the acoustic and articulatory records every eighth pulse dif-

fered f r o m the r e s t in shape and would produce a thicker t r a c e on

the film. In Fig. I-A-7 the synchronization pulses appear at the top

of the spectrogram. The shape of the pulses was chosen s o a s to

speed up the onset charac ter i s t ics of the lamps. The delay between

the onset of a given pulse and the onset of the corresponding photo-

graphic t r a c e was thus rendered negligible. The decay charac ter - - 1 i s t ics of the lamps permitted no f a s t e r r a t e than 50 sec fo r the

pulses to appear distinctly on the film. In view of the relatively slow

movements of the lips and the jaw this was considered quite sufficient.

DISTANCE BETWEEN L I P LAMPS (mm)

STL-QPSR 4/1967

It was c l ea r that the movements of the l ips and the mandible

might not always take place a t right angles with the f i lm path. Often I

such movements a r e a l so t ransversa l . A lamp can accordingly be

displaced horizontally away f r o m i t s ideal ver t ical path, that i s , i along the t ime scale of the f i lm and consequently dis tor t the t ime

relations among the t races . One valuable feature of using pulsed I

instead of continuous t r a c e s i s that such distortions can be corrected

for. Fig. I-A-9 shows an example of the type of record that can be 1

obtained with the present procedures, i ( I

During a typical r e c o r d i q session the subject would be sc2tcd in

a darkened quiet room reading the t e s t words f r o m spotlit sheets of

paper. Three subjects, a l l laboratory employees, were used. Two

of these (subjects A and C) had had no training before a s experi-

mental subjects and were not aware of the purpose of the experi- i

ments. The third ta lker (subject B) was the present wri ter . All 1 t h ree have normal speech and speak a Stockholm variety of Swedish.

They a l l have normal occlusion (Angle, c l a s s I). Talker A has a 1 slightly open bite. They a lso served a s subjects in the experiment

of Fig. I-A-1.

The speech mater ia l s recorded contain the four nonsense words

[I ' b a :bI], [ I J bi:bI] , [I' babbI] , and [I ' b ~ b b ~ ] which were pro-

nounced i n the context of [ s e j - I' jenl ( say again). The ab-

solute level of the speech signal one m e t e r f r o m the t a lke r s J lips 2

was of the o rde r of 70 dB rel . 0,0002 dyn/cm . After about a doz-

en i tems had been read f r o m the l i s t the recording was interrupted

and the ta lker would be "unchained" for a short while. Before each

new recording c a r e was taken to reproduce the standard positioning

s o that the Frankfurt horizontal would be paral le l with the axis of

the c a m e r a lens system. This procedure was repeated until a suf-

ficient number of i tems had been recorded. On the f i lms a given ! sequence of t e s t words was preceded by a recording of the mandible

i n a maximally elevated position s o that i t s location i n relation to

the maxilla could b e inferred. The degree of jaw opening a s used i n

Figs. I-A-1 and I-A-2 thus r e fe r s to the vert ical projection of the

distance of the jaw lamp f r o m i t s position for closed jaws. To iso-

la te the jaw-dependence of the distance between the lamps f r o m i t s

- time

Fig . I-A-9. Example of f i lm r eco rd of a r t i cu la to ry movements . The t r a c e s marked A, B, C and D per ta in to the re fe rence lamp, the upper l ip, lower l ip and jaw l am ps respectively. The ut terance i s [I ' bi:bI] .

STL-QPSR 4/1967

labio-muscular dependence the subjects also opened and closed their I

mouths slowly with a s relaxed lips a s possible. Individual plots of

the separation of the lamps [x - x where xu = x = 01 a s a U - ~ A j' a

function of jaw position indicate linear relationships with propor-

1 tionality constants not markedly different f rom one. The total

scat ter around lines fitted by visual inspection to these plats was

found to be approximately f 2 .5 mm.

F r o m past experience i t is c lea r that the described method com-

pares well with alternative techniques such a s high-speed(19) and

cineradiographic (20) photography a s regards accuracy of measure-

ment. A definite advantage i s that the time variations of individual

parameters can be obtained without a t ime -consuming examination

of individual film f rames . On the other hand, i t shares many disad-

vantages with other articulatory and physiological measurements

which all require long-winded preparations and tenacity on the part

of the subject. . .

1 1

Experiment a1 results

Quality of data --- ---- The technique described above was used to acquire lip and jaw

data on the production of open and close vowels. These data were ob-

tained f rom enlarged tracings of records such a s Fig. I-A-9. The

accuracy with which spatial coordinates can be measured on the film

i s limited chiefly by the enlargement factor and the resolution of the

light pulses. During fast movements the pulses would sometimes

appear weaker and slightly blurred. This difficulty occurred only

occasionally and the e r r o r of a given measurement i s estimated to

be within a fraction of a millimeter in most cases. Since the pulse

ra te was known measurements of t ime coordinates presented no prob-

lem. In synchronizing the t races c a r e was taken to correc t for dis-

tortions and spurious translations of the t ime scales caused by t rans-

versa l movements, faulty midsagittal alignment of lamps etc. The

temporal alignment of acoustic and articulatory records should be ac-

curate within a few milliseconds. The identification of acoustic seg-

ment boundaries was made from mingograms since i t was found that

the segmentation of stop-vowel stop sequences on such records gave

a s accurate result a s that based on wide-band spectrograms.

STL-QPSR 4/1967 13, ! ! I

In these figures comparisons a r e presented between [I' ba:bI] I

and [I' bi:bI] and between [I' babb~] and [I' b~bbI]. The vertical Lines

represent plosions and closures for [b]. F o r words with a close

s tressed vowel dashed lines a r e used to indicate these boundaries.

F o r the words containing [a:] o r [ a ] solid lines a r e used. F r o m

left to right the lines $ b a d for the moments of closure for the f i rs t

[b], release for the f i rs t [b], closure for the second [b], and re -

lease for the second [b]. The t ime reference common to each pair

of utterances i s the beginning of the occlusion for the initial [b]. 1 The time location of a given boundary corresponds to the moment at

which xu - xQ - x. i s equal to i t s average value for this boundary. J

Talker A: [a:] - [i:], Fig, I-A-10 I

The top part of Fig. I-A-10 shows the distance between the lip

lamps, o r x - x in [I' ba : b ~ ] and [I ' bi:bI]. The curves vary U - ~ I , jy

smoothly through the entire words exhibiting maxima of separation

during the vowels and minima during the consonants. The minimum

for the second [b] i s somewhat deeper for [1'bi:b1]. F o r this word

- x . assumes a larger value at the f i rs t release than a t the Xu - XI, J

closure for the second [b] . It i s also larger for [ a :] than for [i:]

at the lat ter moment. A general feature of the curves marked

xu - xA - X. i s that, during the [b] segments, these curves continue J

to decrease t i l l a minimum i s reached a t a point close to the middle

of the segment. Evidently the compressibility of the lip margins

prevents x - x . f rom becoming "clipped" during the closure U - X ~ J

I intervals.

An examination of x . indicates that the mandibular position for J

[a:] i s anticipated already during the occlusion of the f i rs t [b]. F o r

[i:] x . shows but little opening excursion. Returning to i t s position J

for the second [b] x. changes slowly. As a result the curves for J I

[I' b a :bI1 and [I' bi:bI] differ considerably at the beginning of oc- i

clusion of the final [b]. i The bottom part of Fig. I-A-10 shows the curves corresponding to

x - xR which a r e s imilar with respect to the negative minimum val- u ues reached during the [b] segments and the positive maxima at-

tained during the vowels.

STL-QPSR 4/1967

Talker A: [a ] - [I] , Fig. I-A-11

The curves labeled x - x . differ mainly a s regards opening U - x ~ J

for the s t r e s sed vowel. Also the [a] is longer than [I] whereas the

consonant segments have about the s a m e duration. The separation

of the lips is l a r g e r a t the instant of re lease than a t moment of oc- I

clusion (Table I-A-:Ia and b). I In t e r m s of x. the opening ges ture f o r [ a ] s eems to have been

J initiated well ahead of the plosion. This a p p e a r s b be t r u e a l so for

the [I] curve i n which the point of maximum excursion i s reached

somewhat ea r l i e r than that f o r [a]. The movement f r o m the vowel

[a] towards the final [b] progresses slowly. As a resul t the [b] I 1

following [a ] is delayed.

F o r x u - marked differences can be observed between [ ~ ' b a b b I ]

and [IJbIbbI1. The [I] curve begins to approach i t s positive value

f o r the s t r e s sed vowel ea r l i e r than the [ a ] curve which a lso displays

a somewhat deeper minimum for the second [b]. Testing fo r signi-

ficance in the intervals where the curves differ indicates highly sig-

nificant differences i n the regions of both the f i r s t and the second

[b] occlusions. These t e s t s were performed not on curve segments

but only fo r pa i rs of ordinates a t individual t ime coordinates. Since

fewer degrees of f reedom a r e involved i n t e s t s of the la t te r kind we

a r e justified in concluding that not only the tested pa i rs but a l so the I

curves a r e likely to differ significantly in these intervals. I

Talker B: ra: l - ri:l. Fig. I-A-12

Regarding x - xi - x. we note that it i s l a r g e r a t the moments of u J re lease than a t the beginning of closure.

F o r [a:] x . s t a r t s approaching the s t r e s sed vowel position during J

the f i r s t [b] and reaches its value f o r the second [b] considerably

l a t e r than the initiation of the occlusion.

Talker B: [ a ] - [I] , Fig. I-A-13

- x . i s l a r g e r a t the plosions than a t the moments of Xu - J

closure.

I ' b V b b I

40 '- X u - X l - - X j I I I I I TALKER:A-

n

E 30 - I I I Cal - - E - 20- m

I- Z 10- m

Fig. I-A- 1 1 . Comparison of labial movement components for [I' babbI] and [I' bIbbI1. Talker A.

l lJ z 0 a, 0 x 0 -10- 0

I- -20

2 w r W 1 0 - >

0 - s! -10 -

I I I .I

4 I I I I I I 1 1 I I

0 0.5 sec.

r

Xj - I I I I I

m

- -

I I I I I

I I

I

I I

I I I I I

I 1

I I

-

I I

I

I I

Fig . I-A- 13. Compar i son of labia l movement components for [I ' babbI] and [I ' bIbbI] . T a lke r B.

81 ., . , I

STL-QPSR 4/1967 as*

It can be inferred f rom x. that the jaw movement for [a] begins J

during the f irst occlusion and that the return movement i s completed

not until during the final [b] occlusion,

In neither of the figures does talker B show the reorganization

of xu - xA observed for talker A in Fig. I-A- 11. Furthermore, the

differences between the open and close vowels a s reflected in x a r e j

somewhat smaller than for talker A, especially at the moments of

closure for the final [b]' s.

Talker C: [a:] - [i:], Fig. I-A-14

The duration of the consonant segments i s comparable in the two

words. [a:] i s longer than [i:]. There i s again the tendency for the

lip lamps to be more widely separated at the release than at the be-

ginning of closure. I The x . curve for [ Iy ba:bI] exhibits considerable excursion for the

J s tressed vowel. There i s anticipation of the [a:] position during the

initial occlusion and the attainment of the position for the second [b]

occurs at a slow rate a s observed also in the previously examined ut- I terances.

Considering x - xQ we see that this component reaches a value of u

about 7-8 mm during [i:] but about 2 mm for [a:]. In Swedish [ a : ]

is a slightly rounded vowel whereas [i:] is spread. Significance

tests performed a s described above indicate a high probability that

the dashed and solid curves a r e different in the neighborhood of the

[b] releases. The relative timing of curves for [a:] and for [i:] and

for the f inal vowel [I] i s similar to that observed ear l ier for talker

A in [ IJ babbI] and [IYb1bb1]. The solid curve seems to be delayed

from the f irst [b] occlusion onward.

Talker C: [a] - [I], Fig. I-A-15 I I

The short vowels for this talker produce data approximately anal-

ogous to those for [a:] - [i:]. Thus the consonant segments a r c not

markedly different in duration. [a] i s longer than [I]. Again,

- x. tends to be larger at the plosions than at the beginning of Xu - X~ J

closure.

I I

I

0 0.5 sec. I

Fig. I-A-14. Comparison of labial movement componentc [I' ba :b1] and [I' bi:bI] . Talker C .

0.5 sec.

Fig. I-A- 15. Comparison of labia l movement components fo r [I' b a b b ~ ] and [ ' bIbbI) . Talker C .

In x. the motion towards [a] is initiated already during the first J

occlusion and the value for the final [b] is attained only during the

l a t e r half of the occlusive segment.

The functions representing xu - differ significantly i n the r e - [ gions of the [b] re leases . Also i n these curves some of the events

f o r [I] occur e a r l i e r than fo r [a]. I I

Summary ~f resul ts I (1) The coordination of labial and mandibular ges tures i s charac-

tezized by coarticulation. The mandible begins its opening movement

fo r the s t r e s sed vow1 i n [ I ' b ~ ( : ) b ( b ) ~ ] while the l ips a r e still in posi- - tion for the [b]. Similarly i t completes i t s closing movement for the

second [b) a f te r the labial c losure fo r this consonant has been attained.

These effects a r e present for a l l the talkers.

(2) Comparing the words with open and close vowel pa i rs it is seen

that t he re is reorganization of the "labio-muscular" components of

l ip separation x u - This effect is clear ly present only i n Figs ,

I-A-1 1, I-A-14, I-A-15. t (3 ) The distance between the l ip lamps i s not the s a m e fo r plosions

and c losure onsets. It tends to b e l a r g e r at the moment of re leases ,

It is a l so l a r g e r f o r [a:] a t the c losure of [b], (Table I-A-Ia and

I -A-I~J ' I (4) Open vowels a r e longer than c lose vowels in a l l cases . However,

the differences a r e in some c a s e s somewhat sma l l e r than indicated by

the data i n Fig. I-A-2. I

TABLE I-A-Ia. Distance in m m between lamps a t the moment of c losure of [b]. ,

column mean

A

row mean

13

9.5

10.5

10

10

' preceding vowel

-. -- Talker:

A

[ ' a:] [ ' i:]

C'al

[ ' I ]

[ 11

B C

12.5 / 12 15

I 7 I 10 12 i

14

12.5

12

1 8.5 9,5

7

8

10

10.5

STL-QPSR 4/1967 17 ./ TABLE I-A-Ib. Distance i n m m between lamps a t the moment

of plosion of [b] ,

column mean

Following vowel

[ ' &:I [ ' i:] L ' a3

[ ' I]

[ 11

In teg re ta t ion of xo. There a r e two tendencies i n Table I-A-I that -- ------- require a n explanation. F i rs t ly , the separation of the l ip lamps is

l a r g e r a t plosions than a t the beginning of closures. This discrep-

ancy should probably be attributed to the presence of adhesive forces

that make the l ips s t ick together and appear a s soon a s the lips make

contact o r a s the opening of the l ips is begun. On the termination of

these forces discontinuities can sometimes be seen in the individual

lip t r a c e s a t the moments of release. In these oases i t is possible

that oscillations a l so occur a s found in previous investigations (14,21)

Our ea r l i e r experience of high-speed fi lm analysis of the l ips (22)

convinces u s that such adhesion effects a r e by no means uncommon

under normal speaking conditions and they should not be attributed

exclusively to the present use of lamps and the special jaw device.

The accomodation of these facts by the model of Fig. I-A-4 is a-

chieved most simply by supplementing the spring connection the "lip"

m a s s e s by an element of viscuos damping which is introduced a s the

driving forces applied to the l ips change signs and become opening

forces. Two values must be assigned to the constant x of Eq. (2): 0

one for c losure and one for release. In this revised f o r m the pres-

Talker: B C A

14 rk' 15.5

12 13 14.5

14 14.5 16.5

13.5 13.5 14.5

12 '13 14.5

ent l ip model b e a r s a cer ta in resemblance to FlanaganJ s model of

row mean

14.5

13

15

14

13

-

vocal fold vibration (23) in which a friction component is discretely

introduced during vocal fold contact. 1

Secondly, a t the beginning of occlusion x - x. tends to be U - ~ R J

l a r g e r fo r [a:] than f o r the other vowels. This t rend r ecur s f o r

a l l th ree ta lkers . As remarked ea r l i e r Swedish [ a:] is a somewhat

STL-QPSR 4/1967

component x from xu - - x . might leave a residue, x xi J

- xi J

j u whose shape would partly be influenced by adhesion effects and other

mechanical factors, In a f irst approximation it seems reasonable

to interpret these parameters a s primarily labio-muscular, that is,

they reflect the shortening of the lip muscles. When xu - xi > 0

the total separation of the lips will exceed their neutral separation

determined by the position of the mandible. This situation may occur

because the upper lip has been raised, lower lip lowered o r both.

When xu - xi = 0 the separation of the lips i s equal to i t s jaw-de- I

pendent value. When xu - xi < 0 the action of the labial muscles I

make the separation between the lips smaller than when i t is neutral.

This might be caused by the lowering of the upper lip, the elevation

of the lower lip o r both. I

In Figs. I-A-10 - I-A-15 the general course of x - xi i s similar

for all utterances. Positive maxima a r e reached during vowels and

negative maxima during the consonants. The interpretation of these

curves i s straightforward. During the vowels the lip separation i s

wider than it is when neutral, The upper lip may accordingly be

raised and the lower lip depressed. Rounding and spreading, al-

though different mechanisms, produce the same result in this para-

meter a s seen f rom a comparison between the values during [ a:] and

ti:] for talkers A and B. During the consonants xu - xA goes nega-

tive. Inspection of the individual t races for the upper and lower lips

indicates that the lips approach each other to make the contact for

[b] the upper lip being depressed and the lower lip elevated. These

displacements result in a distance between the lips which i s smaller

than the jaw-dependent separation. 1

In Figs. I-A- 11, I-A-14, and I-A-15 which pertain to talkers A

and C there a r e large differences between the timing of the x u -

gestures f rom the initial [b]' s into the stressed vowels. The phase

relation i s such that the close vowel curve leads that for the open

vowel. The timing difference i s large enough to prevent the acoustic

[b] segments from becoming shorter in duration before the open

vowels and thus also to prevent these vowels from becoming still

longer. This point can be further illustrated a s followe. Suppose

that for [I'ba:bI] of Fig. I-A-14 we compute a hybrid version of

- x . whose x. i s that of [IYba:bI] but whose xu - Xu - J J i s that of [I' bi:bI]. What effect would this have on the

STL-QPSR 4/1967 20.

time l ~ c a t i - n -1f thc boundary between the initial [-b)

and [a:] ? If the xo value for [i:] in Table I-A-Ib i s used the duration

of [b] decreases by approximately 45 msec from 120 to 75 msecr

The corresponding operations for [~ 'babbI] ( ~ a l k e r s A and C) gives

a reduction of the duration of [b] by 20 and 25 msec respectivelyr

It seems rather unlikely that the space-time patterns of muscular

shortening a r e basically the same in these cases and that some mech-

anical effect such a s adhesion could account for the differences in I

x - xe. Rather, i t looks a s if the opening of the initial [b] is made U

with the lips when [i:] o r [I] follow but to a larger extent with the

jaw when [a:] o r [a] follow. In the region of the initial [b] the I course of x - x. remains the same for each open close pair.

umXh? J

The component gestures seem to be shaped with respect to the total I movement. Note also that xu - xe exhibits significantly deeper min-

ima for the final [b] following the open vowels (Figs. I-A- 11, I-A-14,

and I-A-15). The deeper minima imply that the lips strive to approach

each other more closely during the closing movement from the open

vowels thus compensating for the large separation caused by the jaw.

There is evidently adjustment of lip activity but not of the movement

of the more massive mandibular structures.

These compensatory modes of articulatory behavior raise the

question concerning the involvement of sensory feedback in the motor

organization of speech production. It i s well known that the control

of human and animal movement i s extensively based on this prin-

ciple (24). It has been shown that reflex action of the lips can be e-

licited by mechanical stimulation (25). This result appears to in-

dicate that the contraction of the orbicularis oris i s under afferent

control. Furthermore an extensive electromyographic analysis of 1 facial muscle activity during speech has recently been undertaken by

ohman (26) who finds the EMG patterns compatible with an inter- 1 pretation in t e rms of eneral neurophysiological principles of feedback.

Moreover, Hosokawa q 2 7 ) reviewing the literature on the sensory in-

nervation of facial muscles reports one study that professes to have

established muscle spindles in the muscles of the lips and in the

region of the oral angle. Consequently movements such a s those ob-

served in the present study a r e probably a priori most correctly anal-

yzed in t e rms of neural circuitry characterized by feedback.

STL-QPSR 4/1967 21.

Summary of interpretations : The p r imary reason why open vowels ---- ----------- tend to be longer than close vowels is that the extent of mandibular

movement is l a r g e r f o r open vowels even during the occlusion of the

initial consonant and that the mandibular off-glide movement f r o m

the vowel into the final consonant p rogresses s o slowly that the con-

tact between the ar t iculators fo r the consonant is delayed i n the con- , text of the open vowel. In the present data the timing of the jaw

movement is such that the la t te r effect is the more pronounced.

However, there is some evidence ( ~ i g s . I-A- 11, I-A- 14, and

I-A-15) that the l ip ges tures proper can be reorganized i n such a 1 way s o a s to compensate fo r the contribution of the jaw movement to

the opening of the l ips for the initial [b]. I

In Eq. (2) xo is one of the determinants of the temporal occur- I rence of r e l ease and closure. Since i t was found to a s sume a l a r g e r

value af te r the rounded [ a : ] than af te r [i:] this difference contri- 1 butes a l so towards reducing the variability of vowel duration that the

timing and the sluggishness of the mandible tend to bring about.

In summary, the data demonstrate the r ea l i sm of the lip-man- I

dible model and the assumptions made initially regarding mandibular

timing and r a t e of movement. I

Discussion

Generality of proposed mechanism

The resu l t s obtained encourage the belief that the relative length-

ening of open vowels that can be observed a lso in the contexts of

[d-d] and [g-g] can be given an explanation s imi l a r to that devel- l oped above fo r [b-b] . To make vowel duration dependent on vowel

opening f o r [ d l and C g ] i t will be necessary to construct art icula- I tory models that incorporate separa te control of the mandible and the

t ip and body of the tongue. So f a r the modeling of ar t iculatory s t ruc-

tu res has not been attempted i n such detail. I I

There i s evidence that the durational variations correlated with

vowel opening appear a l so in vowels followed by [p], [ t ] o r [k].

Before these consonants the acoustic termination of the vowel segment

is usually not determined by the onset of the ar t iculatory closure but

by the offset of phonatory activity. Sound source features en ter a s

STL-QPSR 4/1967 22.

c r i t e r i a of the acoustic end of a vowel a l so before voiceless fricatives.

These c a s e s cal l for a consideration of the temporal relations be-

tween phonatory and ar t iculatory ges ture initiations a s well a s of para-

rreters controlling laryngeal vibration which a r e not contained i n the

present f o r m of the model, e. g. , the p r e s s u r e drop a c r o s s the glott is ,

the position and tension of the vocal folds. I

Without fur ther study it is not possible to c la im that the coordina-

tion mechanism under analysis is a universal mechanism of vowel

and syllable production. Coarticulation on which the durational e f -

fects a r e dependent can be looked upon a s a temporal overlap of ma-

neuve rs towards different but concatenated articulatory goals. By

means of this organization the sys tem manages to speed up the actual-

ization of adjacent sounds without having to speed up the r a t e of move-

ment of the individual component gestures . In view of the allegedly

ra ther slow speed of articulatory movement and of mandibular move-

ment in par t icular we would like to suggest tentatively, however, that

the mechanism studied is likely to be fair ly wide-spread among the

languages of the world. To a cer ta in extent this expectation is rein-

forced by the number of languages for which a durational difference

between open and close vowels has s o f a r been reported, e . g . American

English ( 2 8 ~ 29)J Br i t i sh English (30), Danish (31), German (32)

Hungarian (33), Icelandic (34), Italian (35), Russian (36), Spanish (37 - -

Lappish (38), and Swedish (39). It i s t rue , however, that t he re e r e

one o r two studies that have failed to reveal a difference. F o r Nor-

wegian, fo r instance, vowel duration i n the environment of voiced and (40) voiceless f r icat ives was reported not to vary with vowel opening .

No data for stop contexts were given, however.

Previous explanations I

Previous attempts to explain the durational variability of open and

close vowels have produced two major hypotheses: the "energy ex-

penditure" hypothesis, and the "articulatory distance" hypo+hesis.

According to the "ene r sy expenditure" hypothesis the temporal organiza-

tion of speech sounds is determined by the amount of physiological en-

e rgy that i s consumed i n producing them. During a close vowel m o r e

energy is expended than during an opener one. If the energy p e r vowel

were kept constant [ i] would be shor te r than [ a ] . This explanation

was f i r s t introduced by Meyer (30) and has appeared i n related f o r m

a lso in l a t e r discussions.

STL-QPSR 4/1967

consequences of ar t iculatory timing and r a t e were undesirable. Con-

sequently the fact that the adjustments do occur may indicate that

what the sys tem i s trying to achieve is relatively constant acoustic

duration of each segment in a given position in spite of coarticulation

and other disturbing influences.

Note in this connection the relatively constant and vowel independ-

ent duration of the final [b] in Figs. I-A-1 1, I-A-13, I-A- 14, I-A-15.

It i s a s if a "count-down" fo r the [b] begins a t the moment of closure.

This temporal patternirlg brings to mind the hypothesis of "chain-

reflex" automatisms in motor timing. According to this theory

(which has been crit ized by Lashley (43) and Lenneberg (44)) ,!the

performance of each element of the s e r i e s provides excitation of

the next" ( ).

Conclusions

(1) Evidence f r o m articulatory modeling a s well a s f r o m experimen-

t a l measurements makes i t appear likely that i t is the dynamic be-

havior of thc mandible that gives r i s e to the dependence of acoustic

vowel duration on the degree of vowel opening. 1 (2 ) The universality of this dependence is thus contingent upon the

extent to which the timing and ra te of mandible movement observed

i n the present investigation fo r Swedish a r e typical a l so of other

languages. I (3 ) The resul ts suggest a l so that the motor control of open and

close vowel duration may be character ized by l e s s durational var i-

ability than their acoustic representations.

Summary

In spite cf repeated attacks on the problem, the facts of acoustic

vowel duration a r e still not very well understood. Nor is i t known 1 whether the physiological processes that underlie the variations of 1

vowel duration a r e language-dependent o r universal. Attempts to ex-

plain these facts , i f successful, a r e likely to entail a bet ter under-

standing of the motor organization of the syllable.

STL-QPSR 4/1967

Vowel duration and mandible position

The point of departure of the present study i s the frequently ob-

served correlation between "tongue heightt' and acoustic vowel dura-

tion: The higher the tongue the shorter the vowel. We begin this

paper by examining the open-close dimension of vowels, not in t e rms

of tongue height, but in t e rms of jaw opening. Good agreement with

the traditional vowel quadrilateral i s obtained when the position of

the mandible in long and short Swedish vowels i s plotted against the

classes of front - rounded, front - unrounded,and back-rounded vowels

(Fig. I-A-1). When vowel duration in the context of bilabial stops i s

plotted against mandibular position a positive correlation can be ob-

served (Fig. I-A-2): The lower the jaw the longer the vowel.

& A dynamic model of lip and mandible coordination is constructed

that gives an articulatory interpretation of the dependence of vowel

duration on jaw position. The model in which the lips and the man-

dible a r e represented by damped spring-mass systems permits the

derivation of the course of the midsagittal separation of the lips, 1 s ( t ) for CbVb] syllables. In t e rms of this parameter vowel duration

i s defined a s the interval during which s > 0 and can be determined

from I This equation i s by definition valid only when x ( t ) - xQ(t) - xj(t) -

u - x 2 0. Fo r xU(t) - xQ(t) - x.(t) < 0,s(t) = 0. The t e rm xu i s the 0 J

labiomuscular displacement of a point on the upper lip, x i s the a I

labiomuscular displacement of a point on the lower lip and x . stands J

for the jaw-dependent depression of the lowe r lip. The t e rm x i s 0

the criterion of closure onset or release. This model i s used to com-

pute the hypothetical course of s(t) in [bVb] syllables to which we as-

sign various degrees of vowel o r jaw opening ( ~ i g s . I-A-3 - I-A-6).

The model predicts that vowel duration should increase a s a function

of the mandibular excursion for the vowel provided that certain tem-

poral relations hold between x on the one hand and x and x on the j u a

other: The extent of mandibular movement must be larger for open

vowels even during the occlusion of the initial consonant and/or the

STL-GPSR 4/1967 26.

mandibular off-glide movement f r o m the vowel into the final conso-

nant must p rogress s o slowly that the contact between the ar t iculators

fo r the consonant is delayed i n the context of the open vowel. More-

over, it is shown that the durational variance of open and close vow-

e l s is not necessar i ly a character is t ic of the input control signals to

the model. It is compatible with invariant vowel duration at the level

of control and can b e generated simply a s a resul t of the superposi-

tion of jaw movement on the opening and closing ges tures of the lips. 1

Method. To expose the assumptions Dn which the predictions of the ---- model had to be based data on l ip and jaw movement were collected.

With.the procedure used the l ip and the jaw could be recorded con-

tinuously and i n synchrony with the speech signal. The method in-

volved the use of cylindrical miniature lamps which were attached to

the t a l k e r ' s l ips and to a special device indicating the position of the

mandible. The movement of the lamps during speech was regis tered

by an oscilloscope camera i n which the f i lm r a n perpendicular to the

movement dimension under analysis. In the experiments performed

measurement was made of the ver t ical movements of lamps on the I l ips and the jaw device which were aligned midsagittally.

Results and interpretations. Ten repetitions of four Swedish t e s t --------------- words: [I' b a :bI] , [I ' bi:bI] , [I ' babbI] , and [I' bIbbI] were analyzed

f o r three ta lkers i n t e r m s of the parameters of the model. It was i found that: I ( 1 ) the timing of x the jaw-dependent component, in relation to the

j ' labial components x and x indicates that the coordination of labial

u R and mandibular ges tures i s character ized by coarticulation. The

mandible begins its opening movement for the s t r e s sed vowel i n

[I' bV(:)b(b)I] while the l ips a r e still i n position for the [b] . Simi-

la r ly i t completes its closing movement fo r the second [b] a f te r the

labial c losure fo r this consonant has been attained. These effects

a r e present fo r all the ta lkers . 1 (2) Comparing the words with open and close vowel pa i r s i t is seen

that fo r some of the utterance pa i rs , there is reorganization of the

"labio-muscular" components of l ip separation x u - The effect of

these compensatory adjustments i s to prevent the open vowels f rom

becoming even longer.

STL-QPSR 4/1967

(3) xo i s in general larger at the release than at the closure onset of

a given [b]. This i s assumed to be due to the presence of adhesion

effects. It is larger also for [a:] than for [i:] at the closure of a

following [b]. This i s tentatively attributed to the differences in I

labialization between [ a :] (rounded) and [i:] (spread).

References

( I ) Malmberg, B. : Die Quantitat als phonetisch-phonologischer Beg riff (~und /Le i~z ip j 1 944). I

(2) Delattre, P. : "Some Factors of Vowel Duration and Their Cross- Linguistic Validity", J. Acoust. Soc. Am. - 34 (1 962), 1141-1 143.

(3) It i s known that the loweri of the mandible involves both rota- tion and translation (4. 8 The d ree of jaw opening can be

78) defined in various ways. Poss elt classifies mandibular movements in speech a s "habitual (automatic) opening and closing movements". Although repetitions of these move-

I

ments "do not always coincide exactly they have a fairly char- acteristic main course.. " (Posselt , p. 40). In Fig. I-A-1 the depression of the mandible i s defined a s the vertical projec- tion of a distance along this characteristic path relative to oc- clusion. See also below under section on "Experimental Meth- od and Procedure".

(4) Posselt , U. : Physiology of Occlusion and Rehabilitation o on don 196 6 ) 3 rd printing.

(5) Strenger, F. : "Cephalometric X-Ray Analysis of the Position of the Mandible in the Pronunciation of Swedish Vowels", 0dont.Revy - 7 (1956), 103-117,

I I

( 6 ) F o r further details on the mathematics and physics of the argu- ment of this section see e. g., (7).

I (7) Brown, R. G. and Nilsson, J. W. : Introduction to Linear Systems

Analysis (New York 1962). I

(8) Sievers, E. : Grundzuge der P h ~ n e t i k ( ~ e i p z i g 1893).

(9) Gardner, M, F. and Barnes, J. L. : Transients in Linear Systems ( ~ e w York 196 1) 15th printing.

(1 0) Lindblom, B. : "Studies of Labial Articulation", STL-QPSR 4/1965, pp. 7-9. I I

(1 1) Fujimura, 0. : "High-Speed Motion-Picture Study of the Move- ment of the Lips", MIT, QPR No. 56 (1960), 174-180.

STL-QPSR 4/1967

(1 2) Ohman, S. : "Coarticulation i n VCV Utterances: Spectrographic Measurements", J. Acoust. Soc. Am. - 39 (1966), 151-168.

(13) Ohman, S. : "Numerical Model of Coarticulation", J. Acoust. Soc. Am. - 41 (1967), 310-323. I

I

(14) Lindblom, B. and Bivner, P-0. : "A Method fo r Continuous Recording of Articulatory Movement", STL-QPSR 1/1966, 14-16. ,

(1 5) Jeaffreson, J. W. : "Stress and Rhythm i n Speech", Trans. Philol. Soc. (1 938), 73-91.

(16) Kozhevniksv, V. and C histovich, L. : Speech: Articulation and Perception ( ~ n g l . t ranslat ion f r o m Russian, U. S. Department of Commerce, Washington 1965). I

(17) Pernkopf, E. : Topographische Anatomie des Menschen . I.U. Band: Der Hals (Wien 1952). . I

(18) Anatomical reference line running between a point on the eye orbi t (orbitale) and the most l a t e ra l point in the roof of the bony external auditory meatus (porion).

(1 9) Lindblom, B, and Soron, H. : "Analysis of Labial Movement", J.Acoust.Soc.Am. - 38 (1965)) p. 9 3 5 ( ~ ) .

I (20) Lindblom, B. : "Articulatory Activity i n Vowels", STL-QPSR

2/1964, 1-5. > I

(21) Fuj imura, 0. : "Bilabial Stop and Nasal Consonants: a Motion Pic ture Study and its Acoustical Implications", J. of Speech and Hearing Research - 4 (1961), 233-247.

(22) Lindblom, B.: "Jaw-Dependence of Labial P a r a m e t e r s and a Measure of Labialization", STL-QPSR 3/1965, 12- 15.

(23) Flanagan, J. L. and Landgraf, L, : "Self-Oscillating Source fo r Vocal-Tract Synthesizers", paper A10 presented a t tho 1967 Conference on Speech Communication and Processing, Cambridge, Mass. 1967.

(24) Ruch, T. C. , Patton, H. D. , Woodbury, J . W., and Towe, A. L. : Neurophysiology (1 962).

I (25) Ekbom, K.A., Jernel ius , B . , and Kugelberg, E. : "Peri3ra-l !

Reflexes ", Neurology - 2 (1 952) , 103. I

(26) ahman , S. : "Peripheral Motor Commands in Labial Articulation", a r t ic le 1.B i n this i ssue of STL-QPSR.

I (27) Hosokawa, H. : "Proprioceptive Innervation of Striated Muscles

i n the Te r r i to ry of Cranial Nerves", Tex. Rep. Biol. Med. 19 - (1 96 l ) , 405-464.

(28) House, A. S. and Fairbanks, G. : "The Influence of Consonant Environment upon the Secondary Acoustical Charac ter i s t ics of Vowels", J.Acoust. Soc. Am. - 25 (1953)) 105-1 13.