基于小波包变换和小波阈值消噪的语音特征提取

Phonic character extraction based on wavelet packet transform and wavelet threshold denoising

为了实现强噪声背景下语音信号的特征提取,根据小波变换的多分辨率特性,以及与人耳耳蜗滤波相一致的特性,利用小波包变换,在各语音特征频率段上,提取出包含丰富的非平稳信息的语音特征;并在小波包分解去噪的基础上,构造了模糊阈值函数,利用小波模糊阈值去噪,得到了信噪比较高的语音信号。研究结果表明,小波包变换和小波阈值去噪,较好地消除了强噪声背景下的噪声,并有效地提取出了语音信号特征。

Aiming at realizing phonic character extraction at stronger noise background, wavelet packet Transform was used, and phonic character including plenty nonstationary information was extracted at phonic character frequency segments, based on the multi-resolution feature of wavelet transform, and its consistency with cochlea filtering. Fuzzy threshold function was constructed, and phonic signal with higher SNR was gained, using wavelet threshold denoising. Experimental results show that, noise is denoised and phonic character is extracted availably at a stronger noise background, using wavelet packet transform and wavelet threshold denolsing.