基于维纳–小波分析的语音去噪新方法

A New Speech Denoising Method Based on Wiener Filtering and Wavelet Analysis

小波分析具备多分辨性、低熵性和灵活选择基底的特性,是目前较常用的语音去噪技术之一.但是由于在小波分析去噪中需要设定合适的阈值消除含有噪声信号的小波系数,这会过滤部分有用的语音信号,降低去噪后的语音质量.为了解决这个问题,提出了一种将维纳滤波和小波分析进行结合的去噪新方法.该方法根据维纳滤波具备最小均方误差的特点,可以先过滤掉大部分噪声信号,再进行小波分析二次去噪.仿真实验结果表明,该新方法较小波分析去噪在信噪比和整体质量方面得到较大提升.

Denoising in wavelet analysis sets the threshold to eliminate the wavelet coefficients of the noise, which filters some useful speech signals and reduces voice quality. In order to resolve the problem, a new speech denoising method based on Wiener filter and wavelet analysis is proposed. According to the characteristic of the least mean square error of the Wiener filter, the method filters out most of the noise signal, and then denoises it by the wavelet analysis. Experiments is carried out in simulation, and the results reveal that the method is more improved than wavelet analysis in terms of signal-to-noise ratio（SNR） and speech quality.