基于ICA与GA的语音特征提取方法

Method of Speech Feature Extraction Based on Independent Component Analysis and Genetic Algorithm

为了提高噪声环境中的语音识别率,将独立成分分析(ICA)方法用于语音信号特征提取,并使用遗传算法(GA)将提取出来的高维特征进行选择,最后得到的语音特征被用于基于高斯混合模型的语音识别应用中,并与传统的Mel倒谱系数(MFCC)特征进行比较。实验结果表明基于ICA与GA的语言特征优于传统的MFCC特征。

In order to improve the speech recognition in noisy environment, applies Independent Compo- nent Analysis （ICA） to obtain speech feature extraction. And uses Genetic Algorithm （GA） to select feature from the high-dimensional features. Uses the obtained feature in speech recogni- tion which is based on Gaussian Mixed Model （GMM）. Compared with normal Mel-Frequency Cepstral Cofficients（MFCC）. The experimental results show that the proposed ICA is better than normal MFCC.