一种改进的MFCC参数提取方法

摘要

为了提高语音识别率,提出了一种改进的MFCC参数提取方法.该方法应用小波包变换高分辨率的特点和语音高频加权的功能,在传统MFCC参数的基础上提取了一种新特征参数.新参数能对语音信号频率进行更加精细的划分,能够更稳定地减小频谱失真,且在一定程度上降低了信号的噪声.最后采用高斯混合模型(GMM)进行说话人语音识别,实验表明新特征参数取得了较好的识别率.%In order to improve the speech recognition rate, an improved MFCC parameter extraction method was proposed. The high resolution characteristic of wavelet packet transform and the function of speech high frequency weighted are used in this method, a new feature parameter is extracted on the basis of traditional MFCC parameters. The new parameter can divided speech signal frequency more so- phisticatly, and can stably reduce spectrum distortion, and to a certain extent, can reduce signal noise.Finally, the Gauss mixed model (GMM) is used for speaker speech recognition, and experiment shows that the new characteristic parameters obtaines better recognition rate.