In this paper, we first discuss the questions of feature selection and extraction in the recognition of continuous Chinese speech. And we point out that the pitch information of Chinese voiced sound is necessary to be utilized in continuous speech recognition. Then we propose a segmentation technology of continuous speech based on the pitch information which is extracted from the utterance How real-timely by the pitch-extractor developed by ourselves. Finally, we put forward a recognition technology of Chinese voiced sound based on its spectrum structure. The results of typical experiments are optimistic.
展开▼
