非特定人元音识别的模板假设与实验  

THE TEMPLATE THEORY AND EXPERIMENT OF SPEAKER-INDEPENDENT VOWEL RECOGNITION

作　　者：张红[1] 李治[1] 黄泰翼[2] 

机构地区：[1]西南交通大学,成都610031 [2]中国科学院自动化研究所,北京100080

出　　处：《模式识别与人工智能》1999年第4期431-436,共6页Pattern Recognition and Artificial Intelligence

摘　　要：本文在定量分析了声道长度与共振峰位置的关系之后,结合听觉系统中语音频谱的编码方式,提出了元音识别的模板假设,为解决非特定人语音识别问题进行了初步的探讨。根据这个假设提出了一个可行的元音模板匹配识别算法,通过非特定人语音识别实验证实了元音识别的模板假设的正确性,以及元音模板匹配识别算法的与说话人无关的特性,同时也表明该算法比利用目前主流技术中的倒谱系数的识别方法具有更好的鲁棒性。In this paper, A template matching theory of vowel recognition is proposed, after the relation between the vocal tract length and the location of the formants is discussed theoretically, according to the coding style of the speech in the auditory system, together with a preliminary discussion for resolve the problem of speaker-independent speech recognition. According the theory a feasible template matching algorithm of vowel recognition is presented. The experiments on speaker independent speech samples show that the template matching theory of vowel recognition is correct and that the template matching algorithm of vowel recognition is speaker independent. The results also show that this algorithm is more robust than the recognition method with cepstrum.

关 键 词：非特定人 元音识别 听觉机理 模板匹配 语音识别 

分 类 号：TN912.34[电子电信—通信与信息系统]