An approach of text analysis for Dai speech synthesis

This paper focuses on the approach of the front-end processing of text normalization, words segmentation, Chinese loanwords and Romanization aiming at developing a Dai language speech synthesis system. For normalization, Dai text is divided into standard and non-standard words. This paper discusses problems involved in automatically normalizing Dai non-standard words, and our progress thus far in a rule-based approach to the normalization issue. Non-standard words would be disambiguated and determined its correct pronunciation through the rule-based and context-keyword method. This paper uses the MMSEG algorithm to train the model for words segmentation to improve the accuracy of segmentation. For Romanization, based on the International Phonetic Alphabet of Dai phoneme, Dai character is changed into Roman script via a scheme combined with transcription and transliteration. Due to the existence of a large number of Chinese loanwords in Dai text, we extend the synthetic phoneme of initial and final by analyzing Chinese loanwords' pronunciation rules, and put forward a Romanization method of Chinese loanwords. The experimental results show that the approach proposed in this paper can meet the requirements of the Dai speech synthesis system.