Segment Inventories for Speech Synthesis

Abstract

Speech synthesis may be based on a segmentation of the speech continuum either into simultaneous components or into successive time segments. The time segments may be of varying size and type: phonemes, phoneme dyads, syllable nuclei and margins, half-syllables, syllables, syllable dyads, and words. In order to obtain an estimate of the size of the segment inventory for each type of segment, a phonological study was made of the particular phoneme sequences which occur in English, particularly in relation to the immediate constituents of the syllable (nucleus and margin) and to the syllable. An estimate was also made of the number of prosodic conditions required for each type of phoneme sequence. It was found that in general there is a direct relationship between the length of the segment and the size of the inventory. However, when the borders of the proposed segments do not coincide with the borders of linguistic units, the inventory has to be relatively large. The value of using the various types of segment for speech synthesis is discussed, both for basic research on speech and for practical application to a communication system with high intelligibility.