Disambiguation for polyphones of Chinese based on two-pass unified approach

The paper addresses the issue of Chinese polyphones and the disambiguity approach. Three methods, dictionary matching, language models and voting scheme, are used to retrieve the token's information and then disambiguate the prediction of polyphones. The best precision for these methods achieves 92.72%. Furthermore we proposed the two-pass unified approach to improve the performance with various empirical thresholds. Our approach is superior to the well-known MS Word 2007, and the final precision rate reaches 94.3%. It proves that the proposed approach can resolve effectively text-to-phoneme conversion in Chinese TTS system.