고품위 중국어 TTS(Text-to-Speech) 시스템 구축을 위한 자연어 처리 연구

Abstract

Overall architecture of the Chinese TTS system consists of natural language processing module, grapheme-to-phoneme conversion and prosody prediction module, and speech signal synthesis module. POSTAG/C is a system which combines word segmentation module based on rules and a dictionary with a POS tagging module based on HMM (Hidden Markov Model). The word dictionary was automatically acquired by POS tagged corpus and the system has high portability to serve both GB texts as well as Big5 texts. The word segmenting module is composed of two parts. Input Chinese texts go through a pre-segmenting process at first. The pre-segmenting process splits input texts into short phrases using some special characters that are symbols, numbers, foreign language characters, and mono-character words. Next, we segment in more detailed using four heuristic rules. The two-phase unknown word prediction process extracts candidate boundaries where unknown words possibly occur in a sentence at first. Next, we predict the category of the words with these boundaries using an SVM-based method. The method of our grapheme-to-phoneme conversion is a dictionary-based and rules-based hybrid approach. In this process, our module converts the decided Pinyin strings into more natural forms with standard Chinese phonetic rules.