Automatic Syllable Segmentation of Chinese Speech Based on Types Recognition and Alignment

This paper proposed the algorithm of speech segmentation for Chinese Mandarin and dialects by using the unique syllable structure and types of Chinese. To achieve this goal, The various components consist of Chinese syllables have been analyzed and then the organizational styles of Chinese syllables have been summarized and classified, called as syllable types. For each syllable type, the acoustic model was established by using Gaussian mixture models and hidden Markov models. Finally, the syllable segmentation of Chinese speech was implemented through the process of syllable types recognition and alignment. In the experimental stage, the state numbers of the optimal acoustic models for different syllable types were firstly determined. Further, the stability of acoustic models has been validated on the speech data of Chinese Mandarin and dialects by testing under different experimental environments. The average recognition accuracy can achieve over 93% in these speech data. And the average segmentation accuracy has achieved over 91% on the tolerance threshold of 20ms. By comparing with other algorithms, the proposed algorithm has higher accuracy for Chinese speech segmentation.