Abstract：The wavelet threshold denoising algorithm is simple and its computation is less. However, the discontinuousness of a hard threshold function will cause signal oscillations and a soft threshold function that is too smooth will cause the information loss of high-frequency signal. Based on the disadvantages of two threshold functions, a speech denoising algorithm is developed based on wavelet thresholding, and an improved thresholding function and correction coefficient of revise threshold are put forward. Finally, the Matlab simulation results prove that this method can remove the noise to some extent, reduce the oscillation and distortion of the signals, and retain the characteristics peak information of the original signal. It estimates the original signal very well and improves the speech quality obviously.

Keyword：wavelet thresholding denoising;wavelet transform;thresholding function;