International Symposium on Chinese Spoken Language Processing (ISCSLP 2002)

Taipei, Taiwan
August 23-24, 2002

Automatic Stress Prediction of Chinese Speech Synthesis

Jian-Hua Tao, Sheng Zhao, Lian-Hong Cai

Tsinghua University, Beijing, China

The stress was proved to be the essential links between linguistics and acoustics, and behaves as an important parameter for prosody processing and unit selection in speech synthesis system. In the paper, some acoustical measurements are carried out on F0, duration, silence in order to disclose the relationship between stress and acoustical parameters. The normalized compared acoustic parameters are induced to facilitate the stress detecting from the speech. Furthermore, a rule-learning approach is proposed to predict stress in unrestricted Chinese text. In order to improve the accuracy rate of prediction rules, the most effective linguistic features related to stress are selected according to several experiments. The method is proved to be very successful and has been integrated into our speech synthesis system. We get 86% accurate rate of stress prediction. Further listening tests also show that the expressive force of synthesized speech is improved a lot compared to the systems based on traditional method.


Full Paper

Bibliographic reference.  Tao, Jian-Hua / Zhao, Sheng / Cai, Lian-Hong (2002): "Automatic stress prediction of Chinese speech synthesis", In ISCSLP 2002, paper 82.