DISCOVERY OF TYPICAL PITCH MODELS FROM MANDARIN SENTENCE SPEECH

It is difficult to get accurate pitch variation models by hand, and it is hard to do advanced research based on qualitative and low precise models. To make it possible to acquire prosodic rules more efficiently, clustering analysis is employed to extract pitch models from normalized actual speech, and the result is good enough for future research. In this paper, some background is given at first, and then each stage of the process is introduced in detail. Finally, some experiments are also given.