几种语音识别特征参数的研究  

The Study of Several Speech Recognition Feature Parameters

作　　者：刘雅琴[1] 智爱娟[1] 

机构地区：[1]洛阳师范学院信息技术学院,河南洛阳471022

出　　处：《计算机技术与发展》2009年第12期67-70,共4页Computer Technology and Development

基　　金：河南省自然科学研究项目(2007520035)

摘　　要：语音识别就是研究让机器最终能听懂人类口述的自然语言的一门学科,具有广阔的应用前景。在语音识别系统中,提取语音的特征参数是系统的关键问题之一。文中首先分析了常用的线性预测参数LPCC、梅尔倒谱参数MFCC及其它们的一阶差分△LPCC和△MFCC的原理和实现方法,提取了LPCC+△LPCC和MFCC+△MFCC两种参数。其次,讨论了动态时间弯折DTW识别算法。最后,在Matlab平台上,分别采用LPCC、LPCC+△LPCC,MFCC、MFCC+△MFCC作为特征参数结合DTW识别算法进行实验仿真,结果表明MFCC+△MFCC参数的识别率最高,LPCC的识别率最低。Speech recognition is a subject to study to let the machine eventually be able to understand the natural language of human oral. It has broad application prospects. In speech recognition system , it＇s one of the key problems to extract speech feature parameters. The first,the commonly used linear prediction parameters LPCC, Mel cepstrum parameters MFCC and their first- order differential △LPCC. and/△MFCC, of the principle and method are analyzed. The two parameters of LPCC ＋ △LPCC and MFCC ＋ △MFCC are extracted. The second, the recognition method of DTW is studied. The final, in Mxtlab platform, the experiment is done using feature parameters of LPCC, LPCC ＋ △ LPCC, MFCC, MFCC ＋△MFCC and using recognition method of DTW. The results show that the recognition rate of MFCC＋△MFCC is highest,LPCC is lowest.

关 键 词：特征参数 LPCC 差分倒谱 DTW 

分 类 号：TN912.34[电子电信—通信与信息系统]