Applications of an algorithm for parsing continuous speech into syllables

Abstract

This paper discusses two applications of a previously described algorithm for parsing continuous telephone speech into syllables [D. Kahn, J. Acoust. Soc. Am. Suppl. 1 72, S30 (1982)]. In the first, the parser is used to extract an estimate of the speaking rate, in syllables/rain, of continuous speech input, for use in adjusting rate‐sensitive parameters in speech recognition systems. Since the original algorithm is effective as a syllable parser over a fairly wide range of speaking rates, and since the errors made are of both the missed‐syllable and false‐alarm type, which cancel in the rate‐determination application, a good rate estimate is readily possible. In the second application, the algorithm was specifically tuned for the task of segmenting connected‐digit strings. Since all English digits except “seven” are monosyllabic, a syllabic analysis of connected‐digit strings bears an almost one‐one relation to a segmentation into digits. Further, it has proved possible to adjust the algorithm to quite r...