885095956

Six diphthongs of American English (/au. ai, el, ou, di, ju/) were produced by four midwestcrn American speakers (two małe, two fe-male) at two tempos (slow, fast) with difTering stress (stressed, un-stressed) in two contexts ([b_d], [h_d]). Using a plot of the fundamen-tal frequency and the first three formants derived from linear-prediction-coding (LPC) analysis, the onset and offset of each production was determined. The pattem of formants and fundamental frequency at the onset and offset of diphthongs was uscd to establish a set of parameters that can classify intended productions of the American English diphthongs in varying stress and tempo conditions with an average accuracy of 93%. Results are also presented for diphthong, target-syllabie, and sentence durations. The classification results are dis-cussed with respect to hypotheses conceming the perception of diphthongs.

9:00

8SP5. Auditory-perceptual interpretation of vowels and diphthongs: A progress report. James D. Miller (Central Insi. for the Deaf, 818 S. Euclid Ave., St. Louis, MO 63110)

The auditory-perceptual interpretation of vowels [Miller, J. Acoust. Soc. Am. 85, 2114-2134 (1989)) will be rcviewed briefly and a similar approach to diphthongs, based on spectral glides, will be presented. The relations betwecn acoustic descriptions in the auditory-perceptual space and articulatory descriptions will be outlined. Also, issues relating to the roles of formants versus spectral pattems and the precise meaning of target zones will be presented. The distinction between target zones for vowels produced as steady States as opposed to target zones for vowels produced as spectral glides will be emphasized. Preliminary criteria that may serve to distinguish these will be mentioned. Finally, recent data, which appear to be consistent with the auditory-perceptual approach, will be presented.

noise was added to the filtered vowels (^ = 63.74, p< 0.001). In both experiments, a smali but stable inerease in back-to-front errors in conditions containing cues to filtering was also observed. Results suggest that in these conditions, but not in those lacking cucs, listeners determined which filtered vowels were actual front vowels.

9:30

8SP7. Different discrimination strategies for vowels and consonants. Bert Schouten and Arjan van Hcssen (Res. Inst. for Language and Speech, Univ. of Utrecht, Trans 10, 3512 JK Utrecht, The Netherlands)

In speech perception research, one is often interested in the relation-ship between phoneme Identification and discrimination of stimuli drawn from the same stimulus continuum. If discrimination performance is completely predictable from identifieation, perception is often said to be completely categorical: In both tasks, subjects (can) only use phoneme labels. How should one compare a one-interval identifieation task with a two-interval forced-choice discrimination task in which sub-jeets have to determine the order of the stimuli? Using standard SDT assumptions about optimal processing [D. M. Green and J. A. Swcts, Signal Detection Thcory and Psychophysics (New York, 1974)], it was found that identifieation and discrimination of natural vowels were nearly equivalent, but with natural stop consonants identifieation d was, paradoxically, twice as high as discrimination d. It was concluded that subjects probably use a different strategy for the discrimination of natural stops: They do not subtract the traces of the two stimuli, but the estimated distances between the stimuli and the phoneme prototypes. Such an assumption yields d values that are twice as high as the standard values.

9:45

8SP8. The nonlinear dynamics of categorical perception. Betty Tullcr, J. A. Scott Kelso, Pamela Case, and Mingzhou Ding (Ctr. for Gomplex Syst.. Florida Atlantic Univ., Boca Raton, FL 33431)

8SP6. “Correction” in the perception of filtered vowels. Elizabeth E.

Shriberg (Dept. of Psychol., 3210 Tolman, Univ. of Califomia,

Berkeley, Berkeley, CA 94720) and John J. Ohala (Univ. of Alberta,

Edmonton, Alberta T6G 2E7, Canada)

•

Two studies examined the cffect of cues to channel characteristics on listeners’ perception of low-pass-filtered (1000-Hz) vowels. In experi-ment 1, 85-ms steady-state portions of 11 English vowe!s excised from digitized natural speech were prcccded by a sentence. Following an ulfiltered sentence, filtered front vowels were largcly perceived as back vowels; however, following a filtered sentence, the effect was reduced, and front vowels were “corrccted” at high rates {x* = 108.45, p< 0.001). In experiment 2, the sentence was eliminated and a “masker” was added to the filter-reject region of the stimuli; again, a striking inerease in front-to-back confusions occurred wherf vowels were filtered, and a decrease in these errors occurred when high-frequency

Much research on speech perception ovcr the years has focused on uncovering examples of the nonlinear relationship between acoustics and perception (so-called “categorical perception”). However, little is known concerning the dynamics of this phenomenon. In a variation of the classical categorical perception paradigm, the present experiment explored gradual inereases or decreases in a single acoustic parameter. The resulting pattems of perceptual change showed rich dynamics, in-cluding hysteresis, “anticipation,” a single boundary, and the progres-sion from hysteresis to anticipation over multiple trials. A dynamical system that could account for these perceptual pattems was investigatcd by specifying a potential function that corresponds to the layout of phonetic (attractor) States, and how that layout alters as the acoustic parameter changes. The model reproduces the observed features of the experimental data, and makes further predictions about perception, cur-rently being tested. [Work supported by NIDCD and NIMM.)

10.-00-10:15

Break

10:15

8SP9. The role of multiple acoustic properties in specifying the intemal structure of phonetic categories. Philip Hodgson and Joannę L. Miller (Dept. of Psychol., Northeastern Univ., Boston, MA 02115)

1997 J. Acoust. Soc. Am.. Vol. 89. No. 4, Pt. 2, April 1991

The extent to which a phonetic segment may be specified by morę than one acoustic property was inve$tigated using the trading relation between vowcl duration and closure duration, two properties known to play a role in specifying the voicing comrast for intervocalic bilabial stops. Specifically asked was whether a change in preceding vowel du-

121 st Meeting: Acoustical Society of America 1997