5th International Conference on Spoken Language Processing
The paper gives an overall discussion on problems in Chinese natural speech generation. A Chinese Bi-directional Grammar is developed to suit for Chinese Language understanding and generation. A comprehensive discription about the structure of characteristic network of all ranks in language have been built up. In Natural language generation, text planning is proceeded at first to extract concrete content related to the semantic. Through text organization the internal generation structure is formed. Grammar realization transforms internal structure to natural language. After the text of natural language generated, the next step is to convert the text into speech. We build up a speech characteristic database with speech of 50 thousand phrases and hundreds of pronunciation rules. After recognizing the structure of the input text and abstracting the rhythm characteristics in text, the database gives completely a description from Chinese characters to speech. The whole Chinese character in GB2312-80 can be described to speech. Based on the research all above, a natural speech generation system is established. It can automatically plan and organize the output sentences in natural speech. The synthetic speech has good quality in naturalness and intelligibility.
Bibliographic reference. Fang, Chen / Baozong, Yuan (1998): "The modeling and realization of natural speech generation system", In ICSLP-1998, paper 1034.