10:35-10:50
Break
AH posters will be on display and all authors will be at their posters from 10.50 a.m. to 12:00 noon.
2SP7. Translating pellet positions into constriction features. Stanley Ahalt, Ashok Krishnamurthy, Tzyy-Ping Jung (Electrical Eng., Ohio State Univ., Columbus, OH 43210-1272), Mary E. Beckman, Kenneth De Jong, and Sook-hyang Lee (Ohio State Univ., Columbus, OH 43210-1229)
Tongue constriction features can be estimated from sagittal x-ray pictures of the tongue surface and vocal tract wali. 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 wali is estimated from whole-head seans and a palate tracę. 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 appiied in a replication of Perkell and Nelson’s test of quantal theory using principal component analysis. Quantal theory predicts that the pellet elosest to the constriction site will show least variabilcty, and that the most prccision will be in the dimension perpen-dicular to the vocal tract wali for “quantal” vowels such as /i/. In the warped space, then, the principal component of variation for the rele-vant pellet for these vowels should be paralle! to the x axis. This pre-diction is borne out. (Work supported by the NSF.) 2SP8. An x-ray microbeam study of Hindi vowel articulations. R.
Prakash Dixit and Raymond G. Daniloff (Dept. of Speech Commun. Disord., Louisiana State Univ., Baton Rouge, LA 70803-2606)
Using the x-ray microbeam system of the University of Wisconin at Madison, articulatory movement data were recordcd from pellets at-tached to the tongue, upper and lower lips, lower jaw, and velum while a native speaker of Hindi produced ten orał vowels contained in non-sensc words of the form bVb. The data do not support the traditional descriptions of Hindi vowels given in terms of either tongue height or vocal tract openness. For example, overall position of the tongue dor-sum was higher for /e/ than A/, and similar for /u/ and /o/, and also for A / and /o/. Position of the jaw was lower for /a/ than for /o/. The lips were morę open for /o/, /u/, and /u/ than for /a/, and morę open for /i/ than A/, /e/, and A/. Position of the lips for /e/ and A/was identical. The velum showed somewhat greater elevation for A/ than /i/, and identical elevation for /u/ and /o/. The remaining relations among various structures involved in vowel production were as ex-pectcd. However, overall reiationships among these structures were quite complex.
2SP9. A multisectional representation of the tongue surface based on ultrasound seans for time-varying yocalizatiorw. Marc A. Cordaro (Dept. of Biomedical Eng., Johns Hopkins Univ., 144 New Engincering Bldg., Baltimore, MD 21218), Maureen Stone (National Inst. of Health, Bethesda, MD 20892), Moise H. Goldstein, Jr. (Johns Hopkins Univ., Baltimore, MD 21218), and Michael Unser (National Inst. of Health, Bethesda, MD 20892)
This paper describes a technique for producmg a movie of multiple planes of the tongue surface based on midsagittal and coronal ultra-sound seans of the tongue during time-varying vocalizations. For each scan view, the ultrasound images and acoustic signal of a given vocal-ization are recorded on videotape. The rcsulling sequcnce of video fields is digitized and processed to yield a sequence of two-dimensional coronal tongue profiles. Formant and fundamental frequencies are deter-mined from the acoustic signal segment corresponding to each video field. Data for a number of coronal scan views are recorded with the vocalization repeated for each view. The resulting tongue profile se-qucnces are time warped in a piecewise linear fashion (utilizing char-acteristics of both the ultrasound and acoustic data) to produce a single sequence of multiple time-aligned tongue surface profiles that are dis-played in slow motion or frecze-frame modes.
2SP10. Objective evaluation of vowel pronunciation. Mark J. Bakkum, Reinier Plomp (Dept. of Oto-rhino-laryngology, Free Univ. Hospital, P.O. Box 7057, 1007 MB Amsterdam, The Ncthcrlands), and Louis C. W. Pols (Univ. of Amsterdam, The Netherlands)
Usually pronunciation is evaluated suhjectively by listening. The aim of this research project is to obtain an objective measure of the quality of pronunciation. All 15 Dutch monophthongs and diphthongs spoken by 24 malcs (deaf, foreign, and native Dutch norm speakera) were spectrally analyzed. The digital, real-time analysis system con-sisted of 16 bandfilters according to the critical-band model. Besides level and speaker normalization, time averaging was also appiied so that all voweIs were determined by a single point in a 16-dimensional spec-tral space. By principal components analysis, the number of dimensions was reduced, thus making a plain visualization possiblc. The feasibility of the objective eva)uation has been invcstigated by considering to what extent the spectral information, as expressed in various distance mea-sures, does give an adequate description of subjective judgnients of the vowels, as obtained by expcrienced listeners. The overall scores (aver-aged over 15 vowels, after PC A) show a correlation coefficicnt of 0.97. Correlations per vowel arc lower, but significant in nearly all cases. (Work supported by Netherlands Organization of Scientific Research (NWO) and Institute for the Deaf, Sint-Michielsgestel.j 2SP11. An x-ray microbeam study of the effect of lexical stress on the articulation of intervocallc velar stops. Alice Turk (DMLL, Morrill Hal), Comell Univ., Ithaca, NY 14853-4701)
Previous research on two speakera showed that the closing and opening gestures of the upper lip during the production of intcrvocalic labial stops behave differently, depending on the stress of the following vowel [A. Turk. J. Acoust. Soc. Am. Suppl. 1 88. S56 (1990)]. When the consonant precedes a stressed vowel, its closing gesture (charactcr-ized by the measurement peak_yelocity/vertical_disp!acement) is slowcr than its opening gesture. When the consonant precedes an un-stressed vowel, its closing gesture is faster than its opening gesture. An analysis of velar stops will show whether the pattern observed for labials
1871
J. Acoust. Soc. Am., Vol. 89. No. 4, Pt. 2, April 1991
121 st Meeting: Acoustical Society of America
1871