The Statistical Model of Chinese Word Contours Based on Fuzzy Clustering Method

With the aim of constructing a set of prosodic rules enabling to generate high-quality synthetic speech of Chinese, tone concatenation features were investigated for Chinese words. A statistical model is developed for Chinese word pitch contours based on fuzzy clustering and analysis method. The clustering results shows that word contours are not only depending on the different combination of the tones of the adjacent syllables, but also related nearly to the phonetics of them. More research also shows that word contours are also influenced by different surroundings in the sentence, such as the position of the word, the stress degree of the word, the distance between the current word and the stressed word, the mood of the sentence, etc. The paper studies both the word contour models for isolated di-syllables and the distribution characters of them within different surroundings. It helps us greatly to develop the high-quality prosodic parameters in our TTS system.