A Statistical Approach for Grapheme-to-Phoneme Conversion in Chinese
Zhang Zi · Zhongwen xinxi xuebao · 2002
Grapheme to phoneme conversion is a very important module in a TTS system.How to decide the pronunciation of polyphone characters is a problem that hasn't been solved well in Chinese.By studying a large corpus with corrected pinyin transcription,this paper points out that 41 key characters and 22 key words should be processed.This paper proposes to use the ESC based stochastic decision list to learn pronunciation rules for words ith multi pronunciations.With the generated rules,the error rate for grapheme to phoneme conversion decreased from 8 8‰ to 4 4‰.Annotating corpus for training and testing rules is a labor consuming and time consuming task.This paper proposes a semi automatic approach for corpus annotation, which saves almost half of the working load