The Research On Chinese Putonghua Pitch Period Pattern

The Aircraft Maintenance CBT system is a intelligent platform to make multimedia courseware. It is based on multimedia database and it can make many kinds of multimedia courseware just what you like. Chinese TTS system is a subsystem of CBT, it is aim to change the texts into continuous and natural speech stream.In the speech synthesis system, the prosody model is the bridge of test analyzing and speech synthesizing. So the prosody model that reflects the prosody phenomenon closely is one of the most import components of a speech synthesis system. In order to do that, the paper first reviewed the history of TTS, then introduced some theories about TTS, with emphasis on the research of algorithm of voice activity detection, algorithm of extracting pitch period and pitch pattern analysis.Because the information of tone consists in the final of a single syllable, it is necessary to make a research on syllable detection in continuous Chinese speech. A syllable detection software called"ManualCutTool"has been developed, it roots in traditional algorithms of voice activity detection, but also considers in the requirement of the speech synthesis corpus.Some algorithms of extracting pitch period based on time-domain waveform were studied in this paper, including ACF method, AMDF method and so on. The thesis concluded the advantages and disadvantages of the above algorithms and proposed a new algorithm of extracting pitch period. The new algorithm combines the advantages of ACF which has low too-low rate with the advantages of CSMDSF which has low too-high rate. Lots of experiments show that the new algorithm has a great performance on pitch detection and unvoiced/voiced decision.In addition, the clustering algorithm was studied. This paper studied the disadvantages of traditional K-means clustering algorithm and selected the ISODATA algorithm to analyze the pitch period pattern.The final section is to make a conclusion of the whole thesis and propose some possible improvements on the prosody model generation.