What does sinewave speech tell us about phonetic perception?

Abstract

It has been 35 years since the striking demonstration that intelligible speech can be created by mixing sinusoids that follow the formants of natural speech [Remez et al. (1981) Science 212, 947-950]. A good deal has been written about what sinewave speech (SWS) might reveal about the mechanisms underlying phonetic perception. However, most of the experimental work on SWS has been carried out using well-formed sentences whose intelligibility depends on many factors other than phonetic identification. Results will be reported for experiments designed to measure SWS intelligibility explicitly at the phonetic level. For example, listeners were asked to identify the vowel in SW /hVd/ syllables. Intelligibility was far above chance (55%) but dramatically lower than that of the original utterances. Further tests showed: (1) intelligibility increased substantially when the SW syllable was preceded by a brief SW carrier phrase (CP); (2) tests that alternated trials with and without the SW CP showed that CP enhancement is a real-time effect; (3) SW vowels can be classified using a template-matching algorithm trained on naturally spoken vowels at rates exceeding those of human listeners. Preliminary results will be reported for consonant recognition using SW nonsense syllables. [Work supported by NIH.]