Speech Prosody 2004

Nara, Japan
March 23-26, 2004

A Concatenative-Tone Model with its Parameters' Extraction

Gao Peng Chen, Yu Hu, Yi Jian Wu, Ren Hua Wang

iFlytek Speech Lab, University of Science and Technology of China

This paper presents a novel method to describe concatenativetone in Mandarin with parameters of Fujisaki model. The method is based on an essential assumption that when applying Fujisaki model on Mandarin, the F0 contour mostly depends on how different tone types joint,, We can illustrate the concatenative-tone by tone command, which is a combination of accent commands. The patterns of tone concatenation can be represented by different tone commands. A set of equations are designed to predict the tone commands of natural speech with prosodic information such as tone types and word boundary types, etc as their parameters. Typical categories of tone commands for one-syllable and doublesyllable words have been successfully obtained from F0 contours by solving these equations.. Approach of equation solution is also given in detail in this paper.

Full Paper

Bibliographic reference.  Chen, Gao Peng / Hu, Yu / Wu, Yi Jian / Wang, Ren Hua (2004): "A concatenative-tone model with its parameters' extraction", In SP-2004, 455-458.