The Research Of Prosodic Control Algorithm And Realization For Chinese Speech Synthesis

With the development of science and technology, speech synthesis and speech recognition have been already used in all the fields of society, and have become one of hot-researching fields in human-intelligence, speech signal processing and human-machine multi-medium interaction. In term of speech synthesis, however, Chinese is different from west language family such as on grammar structure, grammar rules, acoustic characteristics and prosodic feature so on. At first, Chinese is language with five tones and different tones are used to express different meanings. Moreover, the tones between both words are influenced each other so as to change their original tones, namely co-articulation. Meanwhile there are short-time breaks in continuous speech, which shows sense of rhythm for spoken person. Prediction, analysis and control on prosodic information such as pitch frequency, time length and magnitude of speech signal are named as prosodic control for Chinese TTS.At present, there are many problems not to solve on prosodic control algorithm for Chinese speech synthesis, so that the synthetic speech quality is relatively low in naturalness and intelligibility. Because the synthetic speech quality has been not reached to the level accepted by user, It is restricted that this technology can be widely applied in the market. As a result, this paper is deep to research the methods of Chinese speech synthesis and its algorithm based on Chinese prosodic knowledge and modern speech processing technology. The main research work is following to:1. According to Chinese acoustic characteristics and prosodic feature such as Chinese acoustic tones and characteristics, Chinese sentence tones and models etc. the author analyzes and researches the relations between prosodic features (pitch frequency, time length and magnitude), stress and break as well as prosodic boundary, proposing the rules of prosodic control for Chinese speech synthesis.2. Analyzing and comparing about prosodic feature and structure of prosodic levels, the author has finished the acoustic analysis of prosodic feature and prosodic boundary. It describes thoughts of constructing models based on prosodic levels, predicting prosodic boundary and prosodic levels control. 3. Determining syllable as the concatenated units and using statistical models based data-driven plus models based on rules, Chinese prosodic model has been constituted for better prosodic control.4. Using PSOLA algorithm, the time length and pitch frequency of the concatenated units are adjusted only to certain scales, which influences the synthetic speech quality. At the same time, the synthesis by means of sentence tones and their prosodic control algorithm has researched in this paper.Using above algorithms and methods, the experiment of Chinese TTS has been accomplished. The experimental results show Chinese TTS and its prosodic control algorithm to be available.