汉语语音识别策略的研究

An Exploration on Speech Recognition Strategy of Spoken Chinese

Based on the demisyllableas the recognition primitive(unit), in this paper, a 3-stage recognition structure is presented. It combines the pattern matching method based on speech production model with the feature rule method based on"knowledge", and permits us to investigate the recognition of spoken Chinese with a new point of view. Via experiments and analysis of utterance manner and spelling pronunciation rule, a series of features used in consonant classification and dicision rules used in various stages were obtained. After adopting some measures, the results obtained from experiment show that the system has fairly good performance. The recognition rate of initial demisyllable is 94.7% and our system reduces the storage and computation by a factor of 10 and 36 respectively from those required for syllable based system.