Abstract
A system for speech synthesis by rule is described employing pseudo-phonemes as phonetic units. After scanning the system construction, the problems of selection and concatenation of phonetic units are discussed. In Japanese speech synthesis systems, the naturalness of synthesized speech depends mainly on the generated pitch contour and spectrum transitions between phonemes. To avoid the problem of coarticulation and to concentrate efforts on developing a set of prosodic rules, we propose a system that generates the spectrum parameter sequences by concatenating the spectrum parameters of pseudo-phonemes which suffer coarticulation effects from preceding and succeeding phonemes. Our system consists of a syntactic analyzer, a prosodic and phonemic symbol generator, a pseudo-phoneme concatenator, a prosodic parameter generator, and an LPC synthesizer.

This publication has 2 references indexed in Scilit: