Search

Articulatory and acoustic variability in the production of five American English vowels was examined. The data were movement records for selected fleshpoints on the midsagittal tongue surface, recorded using the x‐ray microbeam. An algorithm for nonlinearly transforming fleshpoint positions to a new Cartesian space in which the x and y axes represent, respectively, the distance of the fleshpoint along the opposing vocal tract wall and the distance perpendicular to the tract wall, is described. The transformation facilitates a test of Quantal Theory in which variability in the two dimensions is compared over many productions of a given vowel type. The data provide some support for the theory. For fleshpoints near ‘‘quantal’’ constriction sites, the primary variability was in the x dimension (constriction location). The y‐dimension values were more tightly constrained, and the formant frequencies were more significantly correlated with the y values than with the x values. The greater variability in constriction location than in degree was not an artifact of the greater distances traversed in the x dimension between the vowel and constrictions in neighboring consonants, since the pattern was preserved when pellet values were translated to take into account a ‘‘context‐free’’ vowel target (the average values in the context of preceding and following labial consonants). Moreover, the observed correlations between formant values and pellet positions in the two dimensions for [i] and [u] were duplicated in an articulatory‐to‐acoustic modeling test using values for constriction length and cross‐sectional area estimated from the data. The model showed smaller second formant variability in the x dimension than in the y dimension for equal‐sized excursions near the constriction sites for these close vowels, in keeping with the interpretation that speakers exercise less precise control in just the dimensions and regions where quantal stability is available. However, the articulatory pattern was seen not just in vowels which clearly have consonantlike constrictions (the quantal vowels [i], [u], and [@sa ]), but also in nonquantal vowels such as [æ].

Previous studies have suggested two different hypotheses concerning the articulation of stress contrasts in English. Examination of low vowels surrounded by labial consonants generally shows lower jaw in accented as opposed to unaccented syllables, a result originally interpreted as a lower and hence louder vowel [Edwards etal., J. Acoust. Soc. Am. 89, 369–382 (1991)]. On the other hand, a study of nonlow back vowels in alveolar contexts showed higher and backer tongue body in accented syllables, suggesting a more peripheral vowel [de Jong, J. Acoust. Soc. Am. 97, 491–504 (1995)]. This paper reexamines the two hypotheses by looking at both jaw and tongue for both phonemically low and high vowels. Five speakers of Australian English read 20 tokens of two short discourses which placed the name ‘‘Babber’’ or ‘‘Beaber’’ twice each in accented and deaccented positions. Jaw and tongue positions were recorded simultaneously using a magnetometer. On average, the jaw was lower in all accented syllables, although the difference was smaller for the high vowel. At the same time, the tongue body reached significantly higher positions in accented high vowels, and somewhat lower positions in accented low vowels. Thus the two hypotheses can be reconciled by their application to the two different articulators. or

This paper examines the relationship between fundamental frequency(F0) target and sternohyoid (SH) activity in low tones of four different pitch accents and of a phrase boundary in English intonation contours produced at three levels of overall vocal effort. Minimum F0 values for the low targets differed as a function of paradigmatically contrasting tone type and as a function of voice effort level. SH activity level also varied as a function of tone type, in inverse relationship to the F0 value. However, it did not show the same simple relationship to variation in F0 value as a function of overall vocal effort, suggesting a shift in the baseline value due perhaps to concomitant changes in subglottal pressure or to jaw lowering for segmental effect.

Various linguistic roles have been hypothesized for the jaw. In Browman and Goldstein’s ArticulatoryPhonology, it is identified as the common articulator among gestures involving the lower lip, tongue tip, and tongue body, and Goldstein has suggested that this may underly the differentiation in Arabic phonology between oral consonants (labials, dentals, velars) and gutturals (uvulars, pharyngeals, glottals). Keating (1983) proposed that consonants are specified for relatively fixed jaw heights (sibilants higher than stops higher than glides), providing the phonetic basis for sonority sequencing constraints. Macchi (1985) proposed that jaw height during consonants reflects both a passive coarticulation with neighboring vowels, and an active suprasegmental specification (lower in stressed syllables). This paper re‐examines these hypotheses using a corpus contrasting different places of articulation. The hypotheses are supported more or less well, depending on place. For example, Macchi’s hypothesis was borne out for labials, but the jaw was quite low during velar stops in unstressed syllables, usually lower than in the preceding vowel. These results suggest that no linguistic role can be stated straightforwardly for all consonants. Rather, the jaw’s contribution varies in proportion to the distance of the oral articulator from the condyle. [Work supported by the NSF.]

Intonation pattern and syllable duration are thought to be the most salient perceptual cues to phrase stress. However, intonation is an inherently ambiguous cue, since not all English pitch accents involve higher pitch on the accented syllable, and because there are stress contrasts at a lower level, where stress cannot be equated with accentuation. Duration, too, is ambiguous, since it can cue other prosodic contrasts, such as phrasal position. This study examines finer‐grained timing cues to stress. A strain‐gauge system was used to examine jaw opening and closing movements in /popen ayep/ sequences in intonationally accented full‐voweled syllables, unaccented full‐voweled syllables, and completely stressless (reduced‐vowel) syllables produced by four speakers. Measured values for movement duration, displacement, and velocity were consistently larger in accented than in unaccented full‐voweled syllables. However, these differences were nowhere near as large as the differences between full‐ and reduced‐vowel syllables. Reduced syllables also had steeper velocity‐displacement relationships, suggesting a durational difference at the level of gestural dynamics as well. However, no such consistent difference was observed between accented and unaccented full‐voweled syllables. These results support the notion that stress contrasts are not uniformly interpreted in the phonetics at different levels.

Quantal theory claims that sounds which are common in phoneme inventories, such as [i] and [a], are articulated in vocal tract regions where acoustic patterns are relatively insensitive to variation in constriction location. Perkell and Nelson [J. Acoust. Soc. Am. 77, 1889–1895 (1985)] tested this claim using a principal components analysis of x‐ray microbeam data. As predicted, tongue pellets showed most positional variation in an axis parallel to the hard palate in [i] and to the pharynx wall in [a]. An earlier replicative study using [i], [a], and three other vowels [Ahalt etal., J. Acoust. Soc. Am. 89, 1871 (1991)] showed that pellet data are more easily interpreted in terms of relative variation in constriction location and degree if they are transformed nonlinearly into a coordinate space where x and y values express location along and distance from the opposing wall of the vocal tract. The present study examined the relationship between formant values and pellet positions in this transformed space, and found some support for quantal theory. In particular, all good correlations involved positional variation in the y dimension for pellets near the constriction. [Work supported by the NSF.]

Since the jaw’s vertical displacement is largely due to rotation about the condyle, its contribution to a consonant’s constriction should depend on distance from the condyle. An earlier study of one speaker’s productions of English stops showed higher jaw in labials than in velars, as predicted, but highest jaw in coronals. This study investigates more consonant types produced by three speakers each of Korean, French, and Arabic. In general, the hypothesized dependency was supported. Jaw height was greater in labials than in velars, and greater in velars than in uvulars, but contributed nothing to pharyngeals and laryngeals. However, the problematic results concerning coronals were confirmed; the jaw was highest in coronals, particularly [s] and [∫ ]. Moreover, in Korean the jaw was higher in velars coarticulated with a neighboring central vowel [barred eye] than coarticulated with the fronter [i]. The greater contribution to coronal fricatives may be because the lower incisors must be positioned to impinge on the air stream, whereas the greater height for dorsal stops next to [barred eye] may be because the opposing surface (the hard palate) is furthest away from the tongue in that place.

As explanation of interspeaker differences in speech production moves beyond sex‐ and age‐related differences in physiology, discussion has focused on individual vocal tract morphology. While it is interesting to relate, say, variable recruitment of the jaw to extent of palate doming, there is a substantial residue of arbitrary differences that constitute the speaker’s ‘‘style.’’ Style differences observed across a well‐defined social group indicate group membership. Other style differences are idiosyncratic ‘‘habits’’ of articulation, individual solutions to the many‐to‐many mapping between motoric and acoustic representations, which may be set in place in the initial acquisition of motor‐control representations for speech. Perceptual studies of social style differences suggest that perceptibility depends upon the task and upon the hearer’s own group membership. The few studies of idiosyncratic differences suggest that speakers perceive each others’ productions in terms of their own habits. Thus, perceptual compensation for speaker differences must go beyond mere vocal tract normalization. A promising route for describing how listeners compensate for the arbitrary variation of style is an instance‐based (or exemplar) model of speech perception in which the distribution of exemplars is heavily weighted by instances of the speaker’s own productions.

Two recent kinematic studies of syllables with bilabial consonants suggest different dynamic accounts of the durational increase associated with stress: Kelso et al. [J. Acoust. Soc. Am. 77, 266–280 (1985)] interpret regression curves for peak velocity against displacement as evidence that lower‐lip gestures in stressed syllables are less stiff, whereas Nittrouer et al. [J. Acoust. Soc. Am. 85, 1653–1661 (1988)] show that the upper‐lip closing gesture begins at a later phase relative to the jaw opening gesture. Preliminary results from a kinematic study of accented syllables suggest a resolution of the discrepancy. Jaw opening and closing gestures for pop and poppa embedded in sentences that varied the placement of the nuclear pitch accent (sentence stress) were looked at. When accented, gestures were substantially longer, but the velocity‐displacement regression curve showed no decrease in slope, contra Kelso et al. Also, a comparison of observed syllable durations to those predicted by the velocity‐displacement ratios showed that this stiffness index alone cannot account for the greater length. Rather, an accented syllable is longer probably because the closing gesture is later, in accordance with Nittrouer et al. The seemingly contradictory pattern in Kelso et al. may be an artifact of including reduced syllables with unaccented full syllables in a single “unstressed” category. [Work supported by the NSF.]

Tongue constriction features can be estimated from sagittal x‐ray pictures of the tongue surface and vocal tract wall. However, such records cannot be obtained in quantity, making them unsuitable for testing models such as Stevens's quantal theory. The x‐ray microbeam allows larger data sets, but records flesh points rather than surfaces. This paper presents an algorithm for relating the two representations. The vocal tract wall is estimated from whole‐head scans and a palate trace. Pellet positions are then “warped” into a Cartesian space where location along the tract and distance from it are the x and y values. The algorithm has been applied in a replication of Perkell and Nelson's test of quantal theory using principal component analysis. Quantal theory predicts that the pellet closest to the constriction site will show least variability, and that the most precision will be in the dimension perpendicular to the vocal tract wall for “quantal” vowels such as /i/. In the warped space, then, the principal component of variation for the relevant pellet for these vowels should be parallel to the x axis. This prediction is borne out. [Work supported by the NSF.]

Two prosodic categories are particularly interesting for examining the relationship between phonetic units and articulatory timing because they provide a clear contrast in prosodic function: The durational correlates of “sentence stress” mark constituent peaks (nuclear‐accented syllables), whereas phrase‐final lengthening marks constituent edges. When kinematic patterns are contrasted in nuclear‐accented versus unaccented and phrase‐final versus nonfinal syllables, it seems that final lengthening is a local decrease in stiffness, whereas lengthening for accent is a change in the phasing of the final (consonant) gesture relative to the opening (vowel) gesture. When interaction with overall tempo is examined, however, it is clear that the prosodic function cannot be equated directly with the dynamic parameters, since at slow tempo, phrase‐final lengthening affects phasing rather than stiffness. Thus prosodic structure is not the immediate goal of the dynamic representation. Rather, there must be intermediate representations of the phonetic tasks in terms of a local tempo change (for final lengthening) and an increase in the duration of the “sonority peak” of the syllable (for sentence stress). [Work supported by the NSF.]

A study reported at an earlier meeting of this Society examined fine‐grained timing cues to three levels of stress by comparing jaw kinematics in intonationally accented syllables (with full vowels), unaccented syllables (with full vowels), and completely stressless (reduced‐vowel) syllables. For all four speakers in the study, measured values for movement duration, displacement, and peak velocity were consistently largest in the accented syllables and smallest in the reduced‐vowel syllables. This study examines the relationships among these kinematic measures using two different models of the underlying gestural dynamics. The first generated shorter (less stressed) syllables by decreasing the latency of the closing gesture relative to the opening gesture without changing the targeted gestural speed or displacement. The second generated shorter syllables by changing the targeted gestural speed, but decreasing targeted displacement more, so as not to increase the predicted velocities. The first model generated predicted durations, which were closer to the observed distribution of durations among the three stress types, whereas the second generated predicted displacements that were closer to the observed distribution of displacements. Neither generated the observed distribution of velocities, suggesting that a hybrid model is necessary.