Perceptual Vowel Space Research Articles

Perception of normal and spectrally compressed speech by the skin

The development of sensory aids continues to be focused on increasing the number of channels and on alternative processing schemes. The present study compares natural hVd syllables and simple sentences presented with varying degrees of spectral compression [R. Hurtig, J. Acoust. Soc. Am. Suppl. 1 85, S44 (1989)] under vibrotactile and electrocutaneous stimulation conditions. In addition, a comparison of the number of channels used was made. Confusion matrices were constructed from the responses made in a closed set identification task, which followed a subject‐paced exploratory learning task. Multidimensional scaling of the stimuli based on their first three formant values produced the standard representation against which the perceptual data could be evaluated. This congruence analysis graphically illustrates the distortions of the perceptual vowel space. An overall shrinkage of the perceptual space is evident across conditions; however, this effect is not uniform. Warping of the space is observed but in different portions of the space for each of the stimulation conditions.

F0 normalization and talker variability

Fundamental frequency (F0) normalization is an effect observed in vowel perception in which the F0 of the vowel plays a role in perceived vowel quality. It was hypothesized that this perceptual effect involves a process in which hearers adjust a perceptual vowel space when they detect a change of speaker. A seven‐step vowel continuum from “hood” to “bud” was synthesized at two levels of F0 (120 and 240 Hz). These two continua were presented to subjects blocked by F0 level and with items of different F0 randomly intermixed with each other. In the mixed F0 condition, there was a large shift in the identification functions of the vowel continuum as a result of changing F0, while there was no effect of F0 in the blocked F0 condition. Analysis of the reaction time data showed that subjects identified the items more slowly in the mixed condition than in the blocked condition and that this increase occurred for items that were preceded by an item with different F0 and not when successive items had the same F0. These data suggest that hearers adjust a perceptual vowel space when they detect a change of speaker. [Work supported by NIH Training Center Grant No. 50‐315‐26 to Indiana University.]

Cross-Language Differences in the Perception of Natural Vowel Sounds

Recent studies of perceptual vowel spaces deal strictly with steady-state stimuli. In most cases, these stimuli were acoustically similar to vowels in the listeners' native languages. In this experiment, listeners from different native language backgrounds performed triadic comparisons on 12 natural vowel stimuli, produced with falling intonation and stress in the context [bəb__] in order to place listeners in a speech-perception mode. Also, for any listener, the stimulus set included vowels not present in his native language. The stimuli were presented under listener control by a PDP-12 computer. Listeners' judgments were converted to relative difference matrices, supplied with additive constants according to Cooper's best-fit technique, and transformed into scalar products of vectors. The resulting data were factor analyzed by the multimodal PARAFAC procedure (Harshman, this meeting), which provides highly constrained solutions of more theoretical interest than are provided by traditional multidimensional scaling. Factor solutions for 1 through 5 factors were obtained and compared. Results across listeners grouped according to native language have implications for the concept of a “universal” perceptual space.