Modeling Duration and Tonal Coarticulation in a Mandarin Chinesese Synthesis

We present in this paper the results of a duration study and a tonal coarticulation study designed for the concatenative Mandarin Chinese synthesis system developed at the Dresden University of Technology. It is reported that the duration model and the tonal coarticulation model are the two most important components of the prosody control in Mandarin. The material for the study of the two prosody components was extracted from a phonetically and prosodically labeled sentence database uttered by the speaker of the synthesis inventory. This approach ensures a high adaptation of the prosody control algorithm to the speaker-dependent characteristics of the synthesizer’s voice, and turned out to be essential for the improvement of naturalness and user acceptance of the synthesized speech.