Phonetica 39: 337-357 (1982)

Effects of Utterance Position on English Speech Timing

J. E. FLEGE, W. S. BROWN, Jr.

Institute for the Advanced Study of the Communication Processes,University of Florida, Gainesville, Fla. USA

Abstract. 8 speakers of American English produced utterances consisting of one to fivedisyllables ([baba] or (pflpa]). Vowel and stop closure intervals were defined by variationsin supraglottal pressure, sensed through a thin tube inserted in the mouth. Closure wasalways longer for Ipl than Ibl in utterance-medial positions. In utterance-initial position,however, Ib/lengthened more than Ipl so that no duration difference between Ipl and Iblwas observed. Utterance-initial position did not influence vowel duration. In contrast,utterance-final position affected only vowel duration, lengthening both final-syllable un­stressed vowels and stressed vowels in penultimate syllables.

Introduction

The study of speech timing is important not only because temporalspecification is crucial to naturalness and intelligibility [3, II], butalso for the insight it may provide into the underlying organization ofspeech motor control [25]. Researchers have been attempting toidentify and quantify factors which affect the duration of phoneticintervals. By now many such factors have been identified including:(I) 'inherent' duration due to phonological identity [14, 40]; (2)phonetic context and syllable structure [10, 39]; (3) word length andposition within the word [18, 45]; (4) utterance length and positionwithin the utterance [5, 18, 38, 42]; (5) position within the discourse[22]; (6) speaking rate [40], and (7) semantic novelty or frequency ofoccurrence [4, 53].

One factor that has not yet been systematically investigated is theeffect on segment duration of utterance-initial position. Although thearticulatory stricture associated with consonant production is ordinar­ily a well-defined physiological event, stop closure does not provide

338 FLEOE/BROWN, Jr.

clear acoustic landmarks in utterance-initial position. As a result, thetemporal specification of utterance-initial stops has been largelyneglected. However, there is reason to believe that utterance-initialposition may have an important effect on stop closure duration, andthat this effect interacts with the voicing feature of stops.

Voicing Effect on Stop Closure DurationPhoneticians have observed that in many languages the closure

interval of a voiceless stop (such as !p, t, k/) is longer than that of ahomorganic voiced stop (fb, d, g/). So typically does closure durationdistinguish voiced and voiceless stops that it has been considered apossible universal of phonetic implementation [47, 57]. However,recent evidence suggests that temporal specification of the contrastbetween voiced and voiceless stops may actually be a language-partic­ular phenomenon. It seems that in certain languages a voiceless stopmay be produced with a shorter closure interval than its homorganicvoiced cognate [6]; in other languages there may be no durationdifference between voiced and voiceless stops [55], and in still others aduration difference.is restricted to only certain syllable positions [8].

Previous studies of English have shown that !p, t, k! is produced witha longer closure interval than !b, d, g!. Voiced-voiceless differenceshave been noted for stops found: (1) in intervocalic position, bothpreceding and following a stressed vowel [15,28-30,33,34,35,40,44,45, 48, 49]; (2) in word-final position, following both stressed andunstressed vowels [34, 35, 40, 48, 49], and (3) in word-initial position,preceding both stressed and unstressed vowels [30, 34, 35].

However, exceptions have been noted to the general rule thatvoiceless stops are longer than voiced stops in English. Several studieshave shown no duration contrast between voiced and voiceless stops[3, 13, 29, 48, 55]. In other studies, voiceless English stops seemed to beproduced with shorter closure intervals than their homorganic voicedcognates [30, 49, 57].

Such negative or countervailing findings might seem to suggest thatduration differences are not in fact associated with the voicing contrastbetween English stops as is often supposed [12]. However, each appar­ent exception to the more general rule of temporal contrast seems toshare a common attribute. When no duration differences are observed,the stops under comparison seem always to be found at or near thebeginning of an utterance.

Effects of Utterance Position 339

Utterance-initial position may affect stop closure in such a waythat a duration difference between voiced and voiceless stops is notpreserved there. The speech timing model elaborated by LINDBLOMand his colleagues [25, 27, 31] implicitly predicts the lengthening ofstops in utterance-initial position. This model postulates recursiveapplication at the word and phrase levels of a single duration-modi­fying process. Word position is known to influence stop closure dura­tion. Stops are longest in word-initial syllables; somewhat shorter inword-final syllables, and shortest in the medial syllable(s) of multi­syllabic words [27,38,47]. So if the effect of utterance-initial positionon stop closure duration is analogous to that of word-initial position,one would except a stop to be longer in utterance-initial position thanin utterance-medial position. In fact, WESTBURY[58] provides a smallamount of data based on high-speed films indicating that stops areindeed lengthened in utterance-initial position. Moreover, it appearsfrom his data that voiced stops are lengthened proportionally morethan voiceless stops, so that in utterance-initial position voiced stopsare somewhat longer than voiceless stops. Thus, the main purpose ofthis study was to determine if stops are lengthened in utterance-initialposition, and whether such lengthening (if it exists) might differentiallyaffect the duration of voiced and voiceless stops.

U tterance- Final LengtheningIn English phonetic segments are considerably longer in utterance­

final than utterance-medial positions. OLLER [38] refers to this pheno­menon as final syllable lengthening owing to the fact that most of thelengthening observed in his study was confined to segments found inabsolute-final CV(C) syllables. GAITENBY[9], however, regarded thistiming effect as the lengthening of final words. KLATT[16] suggests thatutterance-final lengthening (as opposed to the phrase-final lengtheningobserved at utterance-medial syntactic boundaries) may extend overseveral syllables and thus represent a 'general deceleration of motoractivity at the end of speaking acts' (p. 1212). In fact, a limited ten­dency for lengthening to occur in non absolute-final syllables can benoted for individual speakers of English [21,38] and Swedish [27,31].

Timing studies often reveal a simultaneous influence on segmentduration of several factors. Combined effects are usually smaller inmagnitude than one would expect if each of the several factors inde­pendently influenced segment duration. Phonetic segments seem to

340 FLEOE/BROWN, Jr.

resist shortening beyond some minimum duration [14, 18). Forexample, a 'vowel' already shortened due to its position in the wordmay show relatively little durational modification due to its positionin the utterance [31] or due to the voicing characteristic of a followingstop [14, 40].

There appears to be an interesting interaction between two well­known timing factors which influence utterance-final syllables inEnglish. It has often been observed that vowels are longer before finalvoiced stops (as in bad) than before final voiceless stops (as in bat).

However, several recent studies suggest that this timing effect is con­siderably reduced or not evident in non final syllables [15, 51].

This raises the question of whether the effect on vowel durationattributed to the voicing characteristic of a following stop might notactually result from a differential effect of final lengthening on syllablesterminated in voiced versus voiceless stops. An examination of datapresented by OLLER [38] indicates that vowels in absolute-final evesyllables were lengthened about equally when terminated by voicedand voiceless stops. This suggests an independence - and thus separateexistence - of the stop voicing and utterance-final timing effects. Thisissue deserves further scrutiny, however, so another purpose of thisstudy was to examine these factors in combination. Specifically, wewished to determine if vowels lengthen in the penultimate syllables ofutterances and, if so, whether the voicing characteristic of adjacentstops influence these vowels in the same manner as utterance- medialvowels.

Methods

Subjects and Speech Material8 normal adult female speakers of American English with a mean age of28 years served

as subjects. Each subject produced two randomized lists of utterances at a comfortable rateand intensity level. One list was made up of single phrase utterances consisting of from oneto five tokens of the disyllable [babil] (e.g., 'Baba baba baba'). The other list consisted ofutterances made up offrom one to five tokens of[papil] (e.g., 'Papa papa'). The first vowelof each disyllable was stressed and is referred to as the 'stressed vowel'; the initial and medialconsonants are designated 'prestressed' and 'poststressed', respectively.

Subjects read the speech material from typed pages in a seated position, advancing anindex card to reveal each succeding utterance. Each of the different utterance length condi­tions (one to five words) was represented five times per list, yielding a total of 50 utterancesper subject for analysis (5 lengths x 5 repetitions x 2 lists). An extra utterance placed at thebeginning of both lists was not measured to avoid the familiar 'list series' effect.

Bilabial stops were chosen because articulatory closure at this place of articulation isrelatively easy to define (see below). Nonrounded vowels were chosen to avoid coarticulatoryinteraction with bilabial stop production. Nonmeaningful speech material was used for two

Effects of Utterance Position 341

/I.p

a

a

b

GlOTTAL

PULSING

pbab

Fig. 1. Stop closure was measured from the rise above baseline of supraglottal pressure (a)to the sudden marked decrease in pressure signalling release of constriction (b) . Vowels weredefined as the intervals between stop release and closure, except in absolute-final positionwhere the end of glottal pulsing signalled the end of the 'vowel'.

reasons; (I) to help obviate potential artifacts that might result from a direct comparison ofdifferent meaningful utterances, and (2) to provide an invariant phonetic context for thesegments of interest.

While meaning certainly affects the temporal structure of an utterance or discourse,there are probably more similarities than differences between the kind of speech examinedhere and real speech. Nonsense utterances have been found to manifest prosodic and dura­tional characteristics of meaningful speech [20, 24, 31, 37]. Steps were nonetheless taken toinsure that the speech elicited here resembled meaningful speech as nearly as possible.Subjects were familiarized with the speech material several days before the experiment, atwhich time they were allowed to practice until they could fluently produce the nonsenseutterances without placing contrastive stress or emphasis on any single word [2, 31].

Production of each utterance was carefully monitored during the experiment. Subjectswere instructed to repeat any utterance produced with a noticeable pause, with apparentemphasis on any particular word, or with an intonation contour judged to be atypical ofEnglish declarative sentences.

Instrumentation

Two signals were simultaneously transduced. Voicing (glottal pulsing) was monitored bya throat microphone held against the larynx by an elastic band. Supraglottal pressure (Pio)was sensed through a thin polyethylene tube (inner diameter = 1.78 mm, outer diameter= 2.80 mm) inserted into the buccal cavity at the comer of the mouth. To help preventspurious pressure readings, small holes were bored near the open end of the pressure­sensing tube. The output of the transducer (Statham PM6TC) used to convert air pressurevariations into an analog electrical signal was low-pass filtered (43 Hz cut-off) before being

342 FLEGE/BROWN,Jr.

written out on-line by an optical oscillograph (Honeywell1508A Visocorder) run at a paperspeed ofl5.2 cm/s.

Measurements

The duration offour intervals were measured by hand to the nearest millimeter from theoscillograms and later converted to milliseconds (I mm = 6.25 ms). As illustrated infigure I, stop closure (articulatory stricture) of prestressed and poststressed Ipl and Ibl wasmeasured from the rise above baseline of supraglottal pressure to the sudden markeddecrease in supraglottal pressure associated with release of stricture. Such variations insupraglottal pressure appear to provide a satisfactory basis for defining the closure interval ofbilabial stops. SCHWARTZ[44] reports that measurements of the articulatory closure intervalof bilabial stops simultaneously defined by air pressure and airflow ·coincided'. MULLERand BROWN[36], however, observe that pressure may begin rising in the mouth duringbilabial stop production prior to complete cessation of airflow. SHIPP [46] observed nearsimultaneity between the implosion of bilabial stops and a rise in supraglottal pressure, butnoted a delay of up to 15 ms between release of Ipl (but not Ibl) as determined by visualinspection of high-speed films and the rapid decrease in supraglottal pressure associatedwith stop release. Finally, LUBKERand PARRIS[30] reported duration measurementS forbilabial closure intervals defined simultaneously by mechanical lip contact and supraglottalair pressure variations. Measures derived from the two methods agreed to within 4 ms forIpl and 8 ms for Ib/. In utterance-initial position, however, the labial contact measure greatlyexceeded that derived from air pressure.

A stressed vowel was defined as the interval between release of a preceding stop toconstriction of a following stop. Unstressed vowels were measured from release of a precedingstop to either: (I) constriction ofa following stop, or (2) the final glottal pulse registered bythe throat microphone (in the case of utterance-final vowels).

Results

Factorial analyses of variance with repeated measures on the factorsStress (pre-vs. poststressed), Voicing Up/ vs. /b/), and Utterance Posi­tion (e.g., first word, third word, etc.) were carried out separately forutterances consisting of more than one word. The effect of Stress andVoicing on the duration of vowels and bilabial stops found in the one­word utterances were also examined. F ratios yielded by these analysesare summarized in 'Appendix I'.All three factors were found to affect seg­ment durations, with significant interactions due to Utterance Position.

Stop Closure Duration

Prestressed stops were considerably longer in absolute-initial posi­tion than in utterance-medial position for all 8 subjects. Inspection offigure 2, which displays the mean duration of prestressed stops (alsopresented in 'Appendix 2'), reveals that jbj lengthened more in utter­ance-initial position than /p/. Prestressed /b/ averaged 50 ms (71 %)

Effects of Utterance Position 343

120

80

120

80

~ Ipln

PrestressIbl •

40~------------4 1

'M:>rd

3 4 5

Fig. 2. Mean stop closure duration of prestressed Ipl and Ibl as a function of utterancelength and position. Most data points are based on 40 productions (8 talkers x 5 repetitions);standard deviations are presented in 'Appendix 2'.

longer in utterance-initial position than in utterance-medial positions,while prestressed Ipl was longer by only 29 ms (33 %).

Figure 3 displays the average percentage difference in duration bet­ween utterance-initial and utterance-medial stops in the four multi­word utterances. Note that in every case the lengthening of utterance­initial Ibl exceeded that of Ip/. One effect of unequal lengthening ofIpl and Ibl in utterance-initial position was the neutralization of theduration contrast which distinguished Ipl and Ibl in noninitial posi­tions. Closure duration of prestressed Ipl exceeded that of prestressedIbl by 15 IDS (22 %) in noninitial words. In the first words of utterances- where Ipl and Ibl occurred in absolute-initial position - Ipl averaged10 ms (9 %) shorter than Ib/.

The reversal of the duration contrast between Ipl and Ibl in utter­ance-initial position appears to be the source of the significant inter­action of Voice X Utterance Position in the two-, four-, and five-word

344

100

60

FLEGE/BROWN, Jr.

./plo Ibl

2 3

Utterance length

5

Fig. 3. The percentage by which duration of prestressed stops in absolute-initial positionexceeded the average duration of prestressed stops occurring in utterance-medial positions.

utterances (p<O.OI). Post-hoc tests of paired comparison (Duncan'smultiple range test, a=O.OI) confirmed this interpretation. Althoughprestressed Ipl was often significantly longer than prestressed Ibl inutterance-medial positions, there was either no difference or else asignificant difference in the opposite direction between Ipl and Ibl inutterance-initial position (p < 0.0 1).

The closure duration of prestressed Ipl was significantly longer than that of prestressedIbl in all noninitial words of the four- and five-word utterances. In initial words there wereno significant differences except for the two-word utterances, where Ibl was significantlylonger than Ip/. The duration by which Ibl exceeded Ipl in utterance-initial position wasgreatest in the two-word utterances, where 7 of8 talkers showed a reversal in direction of thecontrast. Only 4 of 8 talkers showed the same reversal for stops initiating three-, four-, andfive-word utterances.

The relatively long duration of utterance-initial stops also seems toaccount for the significant interaction of Stress X Utterance Positionin all the multiword utterances (p<O.OI). The mean duration ofprestressed stops (both Ipl and Ibl) was greater than post-stressed stopsin all noninitial words, a difference which averaged 16 ms (25 %).However, in the first words of utterances the difference was substan­tially larger. Bilabial closure averaged 62 ms (107 %) longer in pre­stressed than posts tressed position because of the lengthening of the

Effects of Utterance Position345

1201

1201 120I Poststress Ipl 0I bl •

I

. j

EE" 80

0

:j/ Wj~~ i40

• ....------ --.--40

{{

q' *',,,

JI120-1120

-------80

-- .... -

Fig. 4. Mean stop closure duration (in IDS) of posts tressed stops as a function of utterancelength and utterance position.

utterance-initial (prestressed) stops. Post-hoc tests revealed that differ­ences between pre- versus poststressed stops were always significantin the first words of utterances, while in non-initial words the differ­ence sometimes did not reach significance (p < 0.0 I). These caseswere: the second word of 'Papa papa'; the second and third words of'Papa papa papa'; and the second words of 'Baba baba baba' and'Papa papa papa'.

Prestressed stops found in the final words of utterances were notlength~ned as were stops found in utterance-initial words. Note infigure 2 that stops were no longer in utterance-final than in utterance­medial words. (Note also that the prestressed stops found in utterance­final words were followed by three phonetic segments.)

Figure 4- displays the mean closure duration of poststressed stops asa function of utterance position. Observe first that the closure durationof poststressed Ip I was consistently longer than that of poststressed Ib I.The closure duration difference between Ipl and Ibl was slightlygreater in final words (22 ms or 39 %) than in nonfinal words (an

346 FLEGE/BROWN. Jr.

•

J~J~-F0

160-1

••

'''J~I''']

Stressed

/• - b

~. 1201

c - p

Unstressed• - b~

120120 0- p

-C!

~ 601601

I 60

,

¥,

,~1

12

16°1

~

,601~III

E

~. 120-1

/120

5 u~g 6°1

0l eo0 :r I,,I ~

,IIII

123{, 123{,5

'M>rd

Word

Fig. 5. Mean duration (in IDS) of stressed and unstressed vowels in /p_p/ and /b_b/environments as a function of utterance length and utterance position.

average 14 ms or 27 %). This may be due to the fact that poststressedIpl was about 11 ms longer in final than in nonfinal words. However,post-hoc tests revealed no significant differences between poststressedstops in final versus nonfinal words for either Ipl or Ibl (p<O.Ol).

Vowel Duration

Utterance-final position affected vowels somewhat differently thanstops, as can be seen in figure 5. Vowels in both absolute-final andpenultimate syllables were longer than comparable vowels found inutterance-medial positions. As might be expected, absolute-final vowelslengthened more (71 ms or 89 %) than did vowels in the penultimatesyllable of utterances (14 ms or 9 %). This unequal lenghtening ofstressed and unstressed vowels in utterance-final words seems to be the

basis of the significant interaction of Stress X Utterance Position in allmulti word utterances (p < 0.01). Post-hoc tests confirmed this inter-

Effects of Utterance Position 347

pretation, showing that: (1) both stressed and unstressed vowels werelonger in final than nonfinal words, and (2) that there were no signifi­cant differences between stressed and unstressed vowels in final words

as there were in non final words (p <0.01).Both stressed and unstressed vowels were consistently longer in a

Ib_bl than in a Ip-pl environment. However, the size of this differ­ence was quite small. Stressed vowels in final words averaged about7 ms longer in a Ib_bl compared to a Ip-pl environment; in non­final words the difference was 9 ms. An ll-ms effect was noted for

utterance-medial unstressed vowels even though a word boundaryintervened between unstressed vowels and the following word-initialstop. (We exclude from consideration here unstressed vowels occurringin absolute utterance-final position.)

Post-hoc tests revealed only a single significant difference betweenvowels in voiced versus voiceless stop environments (the unstressedvowel found in the first word of three-word utterances; p<0.0 1). Therelatively small effect of stop voicing on vowel duration noted here isdue primarily to the fact that 'vowel' duration was measured fromrelease of a preceding stop to implosion of a followings stop. Thus therelatively long aspirated interval normally following release of EnglishIpl was included as part of the 'vowel' interval in Ip-pl environments,whereas a correspondingly smaller 'open interval' following release ofIbl was included as part of the vowel in Ib_bl environments.

Utterance Length

The present data provide little support for the existence of an inverserelationship between the closure duration of Ipl and utterance lengthas noted by SCHWARTZ[44]. Although post-stressed Ipl was slightlylonger (about 6 ms) in one-word compared to multiword utterances,the average closure duration of Ipl remained fairly constant (ca. 70 ms)in multi word utterances.

The absence of an effect of utterance length is probably due to thefact that the stops measured here seem to have approached a minimumduration for bilabial stops [1, 40]. Both word familiarity and numberof prior repetitions of a word are known to decrease segment durationin connected discourse [4]. It is likely that both influences diminishedthe duration of phonetic intervals produced by speakers in this study,for the stops examined here appear to have durational characteristicstypical of fairly rapid conversational speech. Closure of poststressed

348 FLEGE/BROWN, Jr.

Table 1. The mean duration (in s) of disyllabic 'words' ('papa' and 'baba') as a function ofutterance length and utterance position

Word duration IncrementTotalRate0.517

-0.5173.870.415

0.484 0.3820.8994.450.429

0.3760.483 0.3891.2884.650.422

0.3880.3910.471 0.3841.6724.780.430

0.3900.3940.3810.471 0.3942.0664.84

Utterance-medial words are italicized. 'Rate' is expressed in syllables per second. 'Increment'refers to the average increase in total utterance duration with the addition of successivewords.

/p/ and /b/ averaged 62 ms, nearly the same value reported by PORT

[40] for bilabial stops produced at a fast speaking rate, and by UMEDA[52] for stops in a connected discourse. Thus the stops examined heremight be expected to resist further shortening, unlike the relativelylonger bilabial stops examined by SCHWARTZ [44].

Utterance length also appears to have had little effect on word dura­tion. Table I presents the average duration of nonsense disyllables (both'papa' and 'baba') as a function of utterance position and length. Utter­ance-medial words were highly consistent in duration despite differencesin utterance length (three to five words). A consideration of speaking rate(syllables/s) might make the effect of increased utterance length appearto be nonlinear. However, the effect of adding successive words was anincremental increase in total utterance duration which closely ap­proximated the average duration of utterance-medial words (387 ms).

The duration of one-word utterances appears to have greatly ex­ceeded that of words in multiword utterances because of the combined

influence of utterance-initial and -final position [26, 27]. Final wordsin multiword utterances averaged 91 ms longer than medial words.Initial words averaged 37 ms longer than medial words. The averageduration of single-word utterances (517 ms) is just 2 ms greater thanthe sum of the average duration of utterance-medial words (387 ms),plus increments for utterance-final and utterance-initial position.

Discussion

Utterance-Initial LengtheningThe present study appears to reconcile apparently contradictory

findings from previous studies concerning the closure duration of

Effects of Utterance Position 349

voiced versus voiceless English stops. In utterance-medial position itwas found that Ipl closure exceeded Ibl closure. However, as the resultof an unequal lengthening of Ipl and Ibl in utterance-initial position,the duration contrast distinguishing Ipl versus Ibl (both precedingand following a stressed vowel) was neutralized.

We cannot be certain that the 'lengthening' observed here is due toan increased period of bilabial constriction for stops found in absolute­initial position. Stop closure duration was defined in this study solelyby means of supraglottal pressure variations. MULLERand BROWN[36]have shown that supraglottal pressure may begin rising prior to com­plete bilabial constriction (as signalled by a cessation of airflow).Moreover, they note that the extent to which a rise in oral pressureprecedes cessation of airflow may be influenced by the rate of labialclosure. Thus, the relatively longer durations observed for Ipl (29 ms)and Ibl (50 ms) in initial compared to medial positions may possibly bedue in part to a slower rate of closure for stops found in utterance­initial position.

STONE[50] also noted an effect of utterance-initial position on stopproduction. Jaw displacement and velocity were examined duringproduction of Idl in nonsense utterances similar to those examined here([dad~ dad~ dad~ dad~]). The velocity of jaw movement duringproduction of the first stressed syllable in the utterance seemed to belittle affected by the addition of a preceding unstressed syllable ([d~]).However, it appeared that for some talkers the jaw was raised relativelymore during production of [da] when it occurred in absolute-initialposition than when it was preceded by another syllable. More workis clearly needed to determine how utterance position may affectmovement (velocity, acceleration, and displacement) and timingcharacteristics of segmental articulation.

We cannot now offer clear insight into the unequal lengthening ofutterance-initial/pi and Ib/. It is probably worth noting, however, thatthe durational variability of utterance-initial stops was greater thanthat of utterance-medial stops (see 'Appendix 2'). The average stan­dard deviation and coefficient of variation were twice as great forutterance-initial than utterance-medial stops.

A consideration of previous studies suggests that stress may be in­volved in the absence of a duration contrast between Ipl and Ibl inutterance-initial positioq. Data reported by LUBKERand PARRIS[30]indicate that Ipl was longer than Ibl in the word-initial position of

350 FU!OE/BROWN, Jr.

CVCVC disyllables only when stress occurred on the second syllableof the CVCVC. When the first syllable was stressed (i.e., ptipap vs.bdbab) closure of word-initial Ibl was somewhat longer than that of Ip/,just as it was for utterance-initial stops in this study. In our pronuncia­tion of 'I will say CVCVC again' the word say is stressed only whenthe second syllable of the CVCVC is stressed. However, when theftrst

syllable is stressed, no preceding syllable in the carrier phrase appearsto be stressed. A similar reversal can be observed for stops occurringin word-medial position in data reported by STATHOPOLOUSandWEISMER[49]. Again, it is our impression that no stressed syllablepreceded the voiced and voiceless stops under examination.

LEHISTE[23] suggests that at least two stressed syllables are requiredfor the establishment of speech rhythm. If this is true, phonetic seg­ments found at or near the beginning of an utterance may not be asfully incorporated into suprasegmental timing patterns as are similarsegments occurring later in the utterance.

One other factor which might be related to the unequal lengtheningof Ipl and Ib/, again highly speculative, is the laryngeal contrastassociated with voiced versus voiceless s"tops. In intervocalic positiona cyclic devoicing gesture normally serves to insure an interval freeof glottal vibrations during Ipl [60]. Assuming it takes some minimumtime to effect this laryngeal gesture, the labial articulation of Ipl (butnot Ibl) may be prolonged somewhat so that neither preaspiration orexcessive postrelease aspiration occurs [see 57]. The lack of a closureduration contrast between utterance-initial Ipl and Ibl might then berelated to a modification of the laryngeal contrast between Ipl and Iblin utterance-initial compared to utterance-medial position.

Finally, length of utterance-initial Ipl and Ibl may have been affec­ted by the magnitude of supraglottal pressure. MULLERand BROWN[36] propose that afferent feedback from pressure-sensitive mechano­receptors in the supra- and subglottal airways may affect the timingof both glottal and supraglottal events. Talkers have, in fact, beenshown to be sensitive to small pressure variations [32, 41]. Datapresented by SOHWARTZ[44] indicate that Ipl and Ibl (but not 1m/)are produced with longer bilabial closure but smaller peak pressurevalues when whispered than when normally phonated. FLEGE [7]found that supraglottal pressure increased much less rapidly for Ipland Ibl in utterance-initial position than in utterance-medial positions.ROTHENBERG[42] estimated that at least 100-150 ms is required for

Effects of Utterance Position 351

supraglottal pressure to attain levels typical of speech at the beginningof utterances initiated by Ip/. (Note that the medial stops of this studywere shorter than ROTHENBERG'S minimum time constant, while utter­ance-initial stops approximated it.) Thus, it is possible that the releaseof utterance-initial stops is delayed until some minimum pressure hasbeen generated, although to our knowledge no definite link betweenglottal or supraglottal timing and pressure magnitudes has as yet beenempirically demonstrated.

In a pilot study we had a subject produce isolated pay and bay with a pressure-sensingtube inserted in the mouth as described above. In one condition the subject was required togenerate a steady supraglottal pressure of about 5 cm HzO before articulating. In anothercondition he was to approximate the lips before articulating, but without causing pressurein the mouth to rise above atmospheric pressure. (Visual feedback was provided.) The dura­tion oflabial articulation appeared to decrease when initiated from greater than atmosphericpressure. Several other pilot subjects, however, were unable to maintain a constant level ofsupraglottal pressure prior to articulation as required.

Utterance-Final Lengthening

No lengthening was observed for prevocalic stops found in utterance­final syllables. This finding differs from that of OLLER [38], but isconsistent with KLATT'S [16] observation that final position tends toaffect only postvocalic consonants in utterance-final syllables.

Final position was found to increase the duration of vowels in boththe final and penultimate syllables of utterances. Unstressed vowels infinal syllables averaged 71 ms (89 %) longer than unstressed vowels inutterance-medial positions. This is a larger effect than OLLER [38]observed for open syllables (32 ms), but a smaller magnitude oflengthening than he observed for closed final syllables (loa ms).Unlike OLLER [38] we found that stressed vowels occurring in pen­ultimate syllables were influenced by utterance position. Lengtheningof penultimate-syllable vowels, which was noted for all 8 talkers,averaged 14 ms (9 %).

Even though vowels in penultimate syllables were observed tolengthen, the small duration difference distinguishing vowels in aIp pI versus Ib_bl context was not very different there compared toutterance-medial positions. This provides some evidence that utter­ance position and stop voicing influence vowel duration independently.

A better test, however, might be to examine unstressed vowels inpenultimate syllables, for OLLER [38] noted somewhat greater finallengthening for unstressed than stressed vowels for most talkers.

352 FLEOE/BROWN,Jr.

Table II. Duration of phonetic intervals predicted by the KLATT[17] timing model, a5

against mean durations (in ms) actually observed for 'Papa papa papa' and 'Baba babababa'p

apa•p apa•papa

Observed

1201647072 811626673 88178761521,302Predicted

851516166 851516166 85151611331,155

b

aba•baba•baba

Observed

1241695187 681685579 74189561521,272Predicted

801585866 801585866 80158581331,153

lmplicationsfor a Model of Speech TimingThe present data support a hierarchical model of speech timing in

which duration-modifying processes apply at several levels of organiza­tion. The utterance-initial stops of this study appeared to lengthenmore (ca. 40 ms) than did the word-initial (but utterance-medial) stopsexamined by OLLER [38] (which lengthened by only about 20 ms). Arelatively great amount of lengthening would be expected for utter­ance-initial stops if they were lengthened due to both their initial posi­tion in a word and in an utterance.

The present results lead to a suggestion concerning 'inherent' dura­tion in synthesis by rule programs. A recent timing model proposed byKLA'IT [17] was applied to the utterances 'Papa papa papa' and 'Babababa baba'. As can be seen in table II, this model successfully predicts 'many of the timing characteristics observed here. However, the modeldid not accurately predict closure duration differences between Ipl andIb/·

Timing models used in the generation of synthetic speech consistof rules which modify a predetermined input or 'inherent' duration.Inherent durations are based on empirical observations of segmentduration in natural speech. A 5-ms difference between prestressed Ipland Ibl was predicted by setting the inherent duration of Ipl at 85 ms,as compared to 80 ms for Ib/. This underestimated the larger 15-msdifference observed here. For posts tressed stops, the model predicts auniform 3-ms difference between Ipl and Ib/, again underestimatingthe average 15-ms difference observed here and the lO-ms differencereported by UMEDA [52] for connected natural speech.

Effects of Utterance Position 353

Both departures from observed values appear to be the result ofbasing inherent duration on values observed for stops occurring nearthe beginning of an utterance (prestressed stops in eve words embed­ded in 'Say_again'). As we have seen, however, the duration contrastbetween voiced and voiceless stops appears to be neutralized in thisutterance position. Thus the 'inherent' durations used as input for timingrules in synthesis programs should be based on stops in utterance­medial position which are preceded by at least one stressed syllable.

Acknowledgment

Supported in part by NIH grants NS07107 to the University of Florida and NS07100 toNorthwestern University. The authors would like to thank G. CANTER, I. LEHISTE. R.KOENIGSKNECHT. and J. LOCKE for comments on an earlier draft of this article.

Appendix I

Summary of F ratios yielded by separate factorial analyses of variance performed forutterances of different lengths on (A) pre- and poststressed stop closure duration of Ibl and'p/. and (B) stressed and unstressed vowel duration

Number of words in utterance

2

3 45

A. SlopsSubject

(7)9.6··(7) 6.6··(7)14.6**(7)26.0**(7)26.2**Stress

(I) 66.5**(I) 66.9**(I) 161.9··(I) 400.6**(I) 279.6··Voicing

(I) 10.1**(I)3.8**(I) 25.0**(I)70.0**(I) 80.5··Position

(I) 18.9**(2) 36.3**(3) 40.4**(4) 23.3**Voicing x Stress

(I)2.8 (I)9.6**( I)3.3 ( I)5.6**(I)0.5Stress x Position

(I) 51.3**(2) 43.5**(3) 67.6**(4)33.8**

Voicing x Position(I) 15.8··(2)1.4(3)5.6**(4)3.7**

Voicing x Stress x Position(I)3.5 (2)2.3 (3)1.8(4)0.6

Error(149)(301)(455)(609)(753)

B. Vowel.'Subject

(7) 25.9**(7) 21.2**(7)29.5··(7)37.5··(7)36.0**Stress

(I) 69.8**(I) 458.1**( I) 1055.4**(I) 1969.6**(I) 1925.1**Voicing

(I) 0.2(I)8.5**(I)10.8**( I)14.6**(I)24.6**Position

(I) 373.8**(2) 208.2**(3)128.8**(4)73.4**

Voicing x Stress( I) 0.4(I)0.1(I)0.0(I)0.1( I)0.2

Stress x Position(I)89.8··(2)80.0··(3)61.2**(4)52.0··

Voicing x Position(I)1.6 (2)0.5(3)2.4-(4)1.0

Voicing x Stress x Position(I)0.6 (2)2.1(3)1.7(4-)1.0

Error(149)(301)(455)(609)(753)

• p > 0.05; •• p > 0.01.

Appendix 2'"

<.11

The mean duration (in ms) of consonant and vowel intervals as a function of utterance length (one to five words) and utte~ce position'"

2

345

C

VCV CVCV CVCV CVCV CVCV

Ipl

11318378155

(55)(25)(17)(40)

Ibl10618756155

(34)(30)(14)(34)

Ipl

105156637384184-80153

(37)(23)(14)(21)(19)(25)(17) (36)

Jbl13016552866718954157 >'Ij

~(84-)(24)(13)( 17)(13)(29)(14)(43) Qto

Ipl12016470728116266738817876152 ti;

:-(SO) (36)(31)(19)(21)(24)(IS) (20)(16)(23)(17)(39) 0

Ibl124-16951876816855797418956152

ffi

(61)

(26)(14)(20)(18)(28)(16) (22)(15)(26)(16)(41) '-:'Ipl

11716461748316868758816167748717877147

(40) . (24)(15)(18)(19)(26)(13) (21)(15)(23)(13)(19)(15)(28)( 17)(40)

Ibl1241675186691745088 7418054837418256141

(47)(19)(14)(16)(13)(22)(12)(22)(IS)(21)(13)(22)(14)(24)(12)(42)

Ipl

11916662808516970758716668748815868739017076149

(57)(23)(15)(32)(18)(25)(31)(23)(18)(21)(16)(21)(20)(26)(15)(25)(18) (25)(17) (62)

Ibl1221685489661785186 7218052887117054807318056147

(56)(20)(16)(26)(17)(23)(12) (19)(IS)(23)(16)(26)(15)(22)(16)(21)(14)(14)(15) (39)

c = Ipl or Ib/; V = the stressed vowel/al or the reduced unstressed vowell_I.

Effects of Utterance Position

References

355

CANTER, G.: Speech characteristics of patients with Parkinson's disease. Ill. Articula­tion, diadochokinesis, and over-all speech adequacy. J. Speech Hear. Dis. 30: 217-224(1965).CARLSON, R.; ERIKSON, Y.;

emphatic stress patterns in1974).

CARLSON, R.; GRANSTRml, B.; KLATT, D.: Some notes on the perception of temporalpatterns in speech; in LINDBLOM,OHMAN Frontiers in speech communication research,pp. 233-243 (Academic Press, New York 1979).

COKER, C.; UMEDA, N.; BROWMAN, C.: Automatic synthesis from text. IEEE Trans.Audio Electroacoustics AU-21: 293-298 (1973).DANLY, M.; COOPER, W.: Sentence production: closure versus initiation of constituents.Linguistics 17: 1017-1038 (1979).FISCHER-j0RGENSEN, E.: Temporal relations in consonant-vowel syllables with stopconsonants based on Danish material; in LINDBLOM, OHMAN Frontiers in speechcommunication research, pp. 51-68 (Academic Press, New York 1979).

FLEGE, J.: The influence of stress, position, and utterance length on the pressurecharacteristics of English Ipl and Ib/.J. Speech Hear. Res. (in press, 1982).

FLEGE, J.; PORT, R.: Cross-language phonetic interference: Arabic to English. Lang.Speech 21: 125-146 (1981).

GAITENBy,J.: The elastic word. Haskins Lab. Status Rep. Speech Res., SR 2,3.1-3.12(Haskins Laboratories, New Haven 1965).

10 HAGOARD, M.: Abbreviation of consonants in English pre- and post-vocalic clusters)j. Phonet. 1: 9-24 (1973).HUGGINS,A.: Speech timing and intelligibility; in RXQUIN Attention and performance,vol. VII, pp. 278-297 (Erlbaum, Hillsdale 1978).

12 JAKOBsON, R.; FANT, G.; HALLE, M.: Preliminaries to speech analysis (MIT Press,Cambridge 1952).KENT, R.; MOLL, K.: Vocal-tract characteristics of the stop cognates. J. acoust. Soc.Am. 16: 1549-1555 (1969).

14 KLATT, D.: Interaction between two factors that influence vowel duration. j. acoust.Soc. Am. 51: 1102-1104 (1973).

KLATT, D.: Vowel lengthening is syntactically determined in connected discourse.

J. Phonet. 3: 161-172 (1975).KLATT, D.: Linguistic uses of segmental duration in English: acoustic and perceptualevidence.J. acoust. Soc. Am. 59: 1208-1221 (1976).

KLATT, D.: Synthesis by rule of segmental durations in English sentences; in LIND­BLOM, OHMAN Frontiers in speech communication research, pp. 287-299 (AcademicPress, New York 1979).

LEHISTE, I.: The timing of utterances and linguistic boundaries. J. acoust. Soc. Am. 51:2018-2024 (1972).

LEHISTE, I.: Rhythmic units and syntactic units in production and perception. J.acoust. Soc. Am. 54: 1228-1234 (1973).

20 LEH1STE, I.: Interaction between test word duration and length of utterance. WorkingPapers in Linguistics No. 17, pp. 160--169 (Department of Linguistics, The Ohio StateUniversity, Columbus 1974).

LEH18TE, I.: Some factors affecting the duration of syllabic nuclei in English j in DRACH­MANProc. 1st Salzburg Conf. on Ling., pp. 81-104 (Gunter Narr 1975).

2

3

4

5

6

7

8

9

II

13

15

16

17

18

19

21

GRANSTROM,B.; LINDBLOM, B.; RApp, E.: Neutral and

Swedish (Speech Communication Seminar, Stockholm

356 FLEOI~/BRowN, Jr.

22 LEHISTE, I.: Perception of sentence and paragraph boundaries; in LINDBLOM, OHMANFrontiers in speech communication research, pp. 191-202 (Academic Press, New York1979).

23 LEHiSTE, I.: Personal commun. (1972).24 LtBEIWAN, M.; STREETER, L.: Use of nonsense-syllable mimicry in the study of prosodic

phenomena. J. acoust. Soc. Am. 63: 231-233 (1978).25 LINDBLOM,B.: Some temporal regularities of spoken Swedish; in F ANT and TATHAM

Auditory analysis and perception of speech, pp. 387-395 (Academic Press, London1975).

26 LtNDBLOM,B.j LYBERG, B.j HOLMGREN, K.: Durational patterns of Swedish phonology:

do they reflect short-term motor memory processes? (Indiana University LinguisticsClub, Bloomington 1981).

27 LINDBLOM,B.; !tAPP, K.: Some temporal regularities of spoken Swedish. Papers fromthe Institute of Linguistics, No. 21 (University of Stockholm, Stockholm 1973). -

28 LISKER, L.: Closure duration and the intervocalic voiced-voiceless distinction in English.Language 33: 42-49 (1957).

29 LISKER, L.: Stop voicing and voicing in English; in V ALDMANPapers in Linguistics andPhonetics to the memory of Pierre Delattre, pp. 339-343 (Mouton, Paris 1972).

30 LUBKER, J.; PARRIS, P.: Simultaneous measurements of intraoral pressure, force oflabial contact, and labial electromyographic activity during production of the stopcognates Ipl and Ib/. J. acoust. Soc. Am. 47: 625-633 (1970).

31 LYBERG, B.: Some observations on the timing of Swedish utterances. J. Phonet.5:49-59 (1977).

32 MALJi.COT,A.: The effectiveness of intraoral air pressure pulse parameters in distin­guishing between stop cognates. Phonetica 14: 65~1 (1966).

33 MAWCOT, A.: Mechanical pressure as an index of 'force of articulation'. Phonetica 14:169-180 (1966).

34 MALECOT, A.: The force of articulation of American stops and fricatives as a functionof position. Phonetica 18: 95-102 (1968).

35 MAw COT, A.: The effect of syllable rate and loudness on the force of articulation ofAmerican stops and fricatives. Phonetica 19: 205-216 (1969).

36 MULLER, E.; BROWN, W.,Jr.: Variations in the supraglottal air pressure waveform andtheir articulatory interpretation; in LAss Advances in basic research and practice,vol. 4 (Academic Press, New York 1980).

37 NAKATANI, L.; SCHAFFER,J.: Hearing 'words' without words: prosodic cues for wonlperception. J. acoust. Soc. Am. 63: 234-245 (1978).

38 OLLER, D.: The effect of position in utterance on speech segment duration in English.

J. acoust. Soc. Am. 54: 1235-1247 (1973).39 PETERSON,G.; LEHISTE, I.: Duration of syllabic nuclei in English.]. acoust. Soc. Am. 32:

693-703 (1960).40 PORT, R.: Linguistic timing facton in combination. J. acoust. Soc. Am. 69: 262-274

(1981).41 PROSEK, R.; HOUSE, A.: Intraoral air pressure as a feedback cue in consonant produc­

tion. J. Speech Hear. Res.1D: 133-147 (1975).42 ROTHENBERG,M.: The breath-stream dynamics of simple-released plosive production.

Biblthca phonet. 6 (Karger, Basel 1968).43 SCHWARTZ, M.: Influence of utterance length upon bilabial closure duration for Ip/.

J. acoust. Soc. Am. 51: 666 (1972).

44 SOHWARTZ,M.: Bilabial closure durations for Ipl, Ibl, and Iml in voiced and whisperedenvironments. J. acoust. Soc. Am. 51: 2025-2029 (1972).

Effects of Utterance Position 357

45 SHARF,D.: Duration of post-stress inter-vocalic stops and preceding vowels. Lang.Speech 5: 26-30 (1962).

46 SHIPP,T.: Intraoral air pressure and lip occlusion in midvocalic stop consonant produc­tion. J. Phonet. 1: 167-170 (1973).

47 Sus, 1.: Consequences of articulatory effort on articulatory timing; in FANT,TATHAMAuditory analysis and perception of speech, pp. 397-413 (Academic Press, London1975).

48 SMITH,B.: Temporal aspects of English speech production: a developmental perspective.J. Phonet. 6: 37-67 (1978).

49 STATHOPOULOS,E.; WEISMER,G.: The duration of stop consonants; in WOLP, KLATTSpeech communication papers. 97th Meet. Acoust. Soc. Am., pp. 197-201 (AcousticalSociety of America, New York 1979).

50 STONE,M,: Evidence for a rhythm pattern in speech production: observations of jawmovement. J. Phonet. 9: 109-120 (1981).

51 SUBTELNY,J.; WORTH,J.; SAKUDA,M.: Intraoral pressure and rate of flow duringspeech. J. Speech Hear. Res. 9: 498-519 (1966).

52 UMEDA,N.: Vowel duration in American English. J. acoust. Soc. Am. 68: 434-445(1975).

53 UJdEDA,N.: Consonant duration in American English.J. acoust. Soc. Am. 61: 846-858(1977).

54 VANLANCKER,D.; CANTER,G.; TERBERK,D.: Disambiguation of ditropic sentences:acoustic and phonetic cues.J. Spech Hear. Res. 24: 330-335 (1981).

55 WAjSKOP,M.: Segmental durations of French intervocalic plosives; in LINDBLOM,OHJdANFrontiers in speech communication research, pp. 109-123 (Academic Press,New York 1979).

56 WARREN,D.; MACKLER,S.: Duration of oral port constriction in normal and cleftpalate speech. J. Speech Hear. Res. 11: 391-401 (1968).

57 WEISMER,G.: Control of the voicing distinction for intervocalic stops and fricatives:some data and theoretical considerations. J. Phonet. 8: 427-438 (1980).

58 WESTBURY,J.: Aspects of the temporal control of voicing in consonant clusters inEnglish. Texas Linguistic Forum, No. 14 (Department of Lunguistics, University ofTexas, Austin 1979).

59 WESTBURY,J.; NUJdI,S.: An effect of phonetic environment on voicing control mecha­nisms during stop consonants; in WOLF, KLATTSpeech communication papers. 97thMeet. Acoust. Soc. Am., pp. 93-96 (Acoustical Society of America, New York 1979).

60 YOSHIOKA,H.; LOPQVIST,A.; HIROSE,H.: Laryngeal adjustments in the production ofconsonant clusters and geminates in American English. J. acoust. Soc. Am. 70: 1615­1623 (1981).

Received: July 26,1982; accepted: August 10, 1982

J. E. FLEGE, Department of Biocommunication, University of Alabama in Birming­ham, University Station, Birminglulm, AL 35294- (USA)