汉语普通话声母的分类与识别

Classification and Recognition of Chinese (Putonghua) Consonants

本文提出了一种实时高效的汉语声母识别新方法。采用声母二次识别的策略,即先用五个时域特征参量将声母分类,然后用语音分布知识及模板匹配法对类内元素进一步进行处理以最后确定识别结果,这种新方法已应用于基于声、韵识别的汉语声控打字机,它也适合于不认人识别系统,本方崇已在IBM PC/XT、286等机上实现,经训练识别率可达95％以上。

In this paper, a real-time and efficient method of recognizing Chinese consonants is proposed. A method called two-pass processing method was adopted in our system. Consonants were first classified into several categories according to five time-domain features,and then the consonants in a certain category were recognized by using language knowledge and pattern recognition technique. This method has been applied to Chinese speech controlled typewriter based on recognizing consonants and vowels separately. It is also suitable to be used in a speaker-independent speech recognition system. It was realized in IBM PC/XT or 286 and so on, For a speaker trained, the recognition rate is above 95 percent,