【摘要】 本文提出一种包括英语和汉语在内,对有限单词并与说话人无关的语音识别系统。该系统是利用一系列的语音参数来描述待识别的语音信号,并与词汇表中相应的各参考语音多数比较判别。为了增加识别率,把各种语音信号接音素及其顺序分类。对于汉语类的语音信号,则进一步按汉语的“四声”声调划分子类。本文讨论了按此种分类方法生成参考语音参数表的方法和过程。本系统分类比较细,虽然这将导致要求计算机内存容量增加,但由于在同一语音类别中参考词汇数目相对地减少,这对提高识别速度和识别率均有利。更多还原

【Abstract】 A speeker-independent speech recognition system for limited isolated English and Chinese words is presented in this paper. The distance between the features attracted from the unknown word and the reference templates are calculated. To increase the recognitipn fate, the clustering technique according to phoneme sequence is used. The subgrouping for Chinese words is described based on the "four tones of Chinese pronunciation". The procedure for creating a reference vocabulary according to the strategy mentioned above is also presented in this paper. Although the memory size for this system is larger because of the increasing sufegroups, it will Have higher speed and accuracy for speech recognition.更多还原