Grapheme-to-phoneme conversion in Chinese TTS system

Abstract

Phonetization is an important component in Chinese TTS systems. However, the polyphonic characters make this problem more complex. The paper reports a study on the relation between Chinese characters and their pronunciation, proposes the solution to the disambiguation of polyphonic characters, a dictionary-based method, and a rules-based method. In the rules-based method, we use the statistical decision list method. The phonetization plan has been proved effective experimentally. Improvements in the accuracy of polyphone phonetization are mostly over 10%.