Phoneme Grouping for Speech Recognition

Abstract

A program for associating a given segment of speech with a phoneme group, such as a vowellike segment, fricativelike segment, etc., is described. Some disadvantages of the presently accepted grouping of phonemes into stops, fricatives, vowels, etc., are discussed. Grouping of phonemes used by Sanskrit grammarians seems to be more appropriate for computer speech recognition. A grouping of phonemes into several nonmutually-exclusive sets that combine the concepts of the presently acceptable grouping and those of the Sanskrit grammarians is proposed. This grouping is more suitable in that it permits algorithmic definitions of vowel-like sounds, etc., in terms of easily obtainable acoustic parameters—namely, intensity and zero crossings. Some results obtained using the suggested algorithm are presented.