基于非齐次隐马尔可夫模型的特定人元音的识别方法  

Chinese vowels recognition method based on non-homogeneous hidden Markov model

作　　者：陈立伟[1] 赵春晖[1] 白玉[1] 孙岩 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2006年第2期296-300,共5页Journal of Harbin Engineering University

基　　金：黑龙江省自然科学基金资助项目(F2004-09)

摘　　要：针对特定人汉语元音的语音识别,提出一种基于非齐次隐马尔可夫模型的识别方法.该方法首先提取声道频率响应作为特征参数,然后建立非齐次隐马尔可夫模型来更为精确地刻画真实的语音现象,接着进行语音识别实验,并与齐次隐马尔可夫模型进行比较.实验结果表明该方法可以使特定人的元音的识别率达到98.73%,明显改变了识别系统的性能.该方法具有很好的理论研究前景和实际应用价值.Introduced is a new speech recognition method, which is aimed at recognizing Chinese vowels and based on a non-homogeneous hidden Markov model （HMM）, First extracted were the frequency responses of the vocal tract as the characteristic parameter, and then a non-homogeneous HMM was established to describe the real characteristics of speech more reasonably, The model resulting from this method was used in speech recognition experiments and compared with a homogeneous HMM. Results show this method of vowel recognition rate in speaker dependent mode can reach 98, 73 %, a distinct improvement over homogenous HMMs, This method provides good theoretical background and practical application value,

关 键 词：非齐次隐马尔可夫模型 声道频率响应 语音识别 元音 

分 类 号：TN912[电子电信—通信与信息系统]