基于小波变换语音组合特征参数的提取

Combination Feature Extraction Based on Wavelet Transform in Speech Recognition

本文首先利用小波变换,借鉴MFCC参数提取了新的参数DWTC,然后又把DWTC参数和LPCC参数结合起来,形成一种组合特征参数DWTLPC,该参数既反映了声道响应特性,又克服了短时傅立叶变换在分辨率上的局限性。最后,在Matlab平台上,分别采用LPCC+△LPCC、MFCC+△MFCC、DWTC、DWTLPC作为特征参数,采用DTW识别算法进行实验仿真,结果表明DWTLPC参数的识别率最高。

In this paper,first of all,a new DWTC feature parameters is extracted using wavelet transform and referencing MFCC.Second, DWTC parameter and LPCC parameters are combinated and the DWTLPC Combination parameters is extracted,this parameters reflects the response of channel, but also overcomes limitations in the resolution of short-time fourier transform.Finally,In Matlab platform, the experiment is done Using feature parameters of LPCC＋ △ LPCC, MFCC＋ △MFCC, DWTC,DWTLPC and using recognition method of DTW. The result shows that the recognition rate of DWTLPC is highest.