An exemplar-based model of silent-center syllable perception
Stephen Winters, Keith Johnson · The Journal of the Acoustical Society of America · 1998
Rakerd and Verbrugge [J. Memory Lang. 26, 558–563 (1987)] suggested that the perception of silent-center syllables—CVC syllables with the middle 60% of the vowel edited out and replaced with silence—had to involve some form of speaker normalization. Crucial to their argument was their subjects’ ability to perceive cross-gender hybrids of such syllables. Rakerd and Verbrugge reasoned that even modest accuracy in perceiving these hybrids entailed some sort of normalization of the disparate acoustic qualities of male and female voices in order to derive one perceptual whole. Rakerd and Verbrugge’s study has been replicated in order to test the ability of a computationally implemented exemplar model to perceive silent-center syllables. Exemplar-based modeling utilizes no ‘‘speaker normalization’’ as such, but, nonetheless, the model in this study approximated the performance of human subjects in perceiving silent-center syllables correctly. The model’s performance on the important ‘‘hybrid’’ case was almost identical to subjects’ responses. These results suggest that the human ability to perceive silent-center syllables does not necessarily entail some process of speaker normalization in speech recognition. [This material is based upon work supported under a National Science Foundation Graduate Fellowship and NIDCD Grant No. R29-DC01645-06.]