论文部分内容阅读
汉语语音识别中连续大词汇量的语音识别率较差。若能把连续大词汇量的语音进行实时自动切分为单个音节,便可提高系统的识别率。如何做到对语音识别中音节的自动切分,首先需找出汉语语音音节的特征。本文综合了当前对汉语音节特征的研究成果,通过深入地比较分析,系统地给出了汉语语音音节的功率谱特征和时域特征,为汉语语音音节的自动切分提供算法依据,对提高连续大词汇量语音的识别率有重要意义。
Chinese speech recognition continuous large vocabulary speech recognition rate is poor. If the continuous large vocabulary voice real-time automatic segmentation into a single syllable, you can improve the system’s recognition rate. How to do the automatic segmentation of syllables in speech recognition, we must first find out the characteristics of Chinese speech syllables. In this paper, the current research results on the characteristics of Chinese syllables are summarized. Through comparative analysis, the power spectral features and time-domain features of Chinese syllables are systematically given, which provides an algorithm basis for the automatic segmentation of Chinese syllables. Large vocabulary recognition rate of speech is important.