Use of Cochlea Patterns in Speech Analysis
J. Ryland Mundie, Thomas J. Moore · The Journal of the Acoustical Society of America · 1969
Displays of basilar-membrane displacement generated by employing speech inputs to an electronic model of the inner ear were studied. Included in the sample of speech inputs were nine vowels, each spoken by four speakers (both male and female), and 22 consonants, all spoken by the same speaker. All consonants were uttered as a consonant-vowel pair. Five of the consonants appeared with each of four vowels, four appeared with each of two vowels, and 13 appeared with one vowel. It was found that distinctive features in the generated surfaces could be used to identify the various consonants and vowels employed. With respect to the vowels studied, it was possible to separate and identify the individual speech sounds across speakers. The most important cues for identifying the various speech sounds were found to be those located in the time domain.