1983 Volume 4 Issue 2 Pages 97-106
An approach to speech synthesis of Thai which is a tone language is presented. The stored data are presently cepstral parameters of consonant-vowel-(CV-), vowel-consonant-(VC-) and vowel (V) demisyllables. Speech is produced by concatenating stored cepstrum-coded demisyllables, providing pitch contour for each of monosyllabic sounds and synthesizing the speech through the log magnitude approximation (LMA) digital filter. Interpolation at boundaries of demisyllables is linear. For each tone, the f0 contours are simply synthesized from the information of a rudimentary contour specified by a third order polynomial in the domain of pitch period. The details on determining a rudimentary f0 contour for each tone, segmentation for demisyllable library, and rules for concatenation and pitch assignment are described, along with experimental evaluation tests and discussions.