基于WT-SVD的语音信号双层滤波方法

Method for Double-layer Filtering of Speech Signal Based on WT-SVD

针对语音信号中存在的信号去噪问题,提出基于小波变换-奇异值分解(WT-SVD)的语音信号去噪方法。先利用小波变换(WT)过滤掉含噪语音信号,然后进行奇异值分解去噪。针对传统的小波阈值函数存在阈值选择困难,引入了指数型的修正系数,降低小波系数误差;同时,提出了基于奇异值方差法,用来确定最佳奇异值的个数,使语音信号的降噪效果更好。实验结果表明,该方法简单易行,比单一的SVD方法和WT方法去噪效果更佳,有效地消除了语音信号中的噪声,同时也避免了信号的失真,显著地提高了语音信号的信噪比。

In view of the problem of signal denoising in speech signals,a speech signal denoising method based on wavelet transform-singular value decomposition（ WT-SVD） is proposed. The wavelet transform（ WT） is used to filter out noise signal,and then this signal is denoised by singular value decomposition.Aiming at the problem of difficult threshold selection of traditional wavelet threshold function,the exponential correction coefficient is introduced to reduce the wavelet coefficient error. And a singular value variance method is proposed to determine the number of the best singular values,so that the noise reduction effect of the speech signal is better.The experimental results show that the WT-SVD-based denoising method is simple and easy to perform,has a better denoising effect than a single SVD method and WT method;this method can effectively eliminate the noise in the speech signal and avoid signal distortion,and significantly improve the signal to noise ratio of speech signal.