The Research Of Speech Denoising Algorithm Based On Wavelet Transform

In the process of speech communication, the voice of the both sides will be interfered by other speakers, car whistle noise, thus increasing the difficulty of the receiver to identify voice messages. How to reduce the impact of environmental noise effectively; improve the audibility of the speech, and the human need to solve the real problem. With the rapid development of science and technology, the study of the voice signal denoising research is also in constant progress, and has been applied in many fields.In order to satisfy the needs of the people of the voice signal denoising, the wavelet transform theory is introduced into the signal processing technology. It has a time-frequency local refinement.multi-resolution analysis of the unique performance, which determines its advantages of voice signal denoising is unsurpassed.This article prospects from the speech signal denoising and analysis the development of the wavelet transform, the basic theory and characteristics firstly. Followed by the wavelet transform modulus maxima reconstruction filter, spatial correlation filtering threshold denoising three categories commonly used filtering methods are summarized; Then focus on the wavelet threshold denoising method, and decomposition scale for the algorithm, wavelet basis function, threshold and threshold function affecting denoising effect parameters detailed discussion and experimental simulation, have chose the more suitable parameters of the speech signal processing factor.Based on the wavelet parameters selection experiment and analysis of test data concluded:the algorithm of the selection of threshold function directly affects the wavelet reconstruction signal precision and continuity. Although the traditional hard threshold function application is simple and convenient, but its result in denoising threshold with discontinuous function effect is not ideal, easy to make the filtered signal after filtering appear Pseudo-Gibbs phenomenon, while the soft threshold function is continuous, its filtering a fixed difference in the coefficients after the original coefficients, likely to cause the deletion of the high frequency information. This paper mainly study the hyperbolic threshold function, law of the threshold function and Garrote threshold function based on introducing decomposition scale. Then put forward two new algorithms based on Garrote function, and derived from mathematical theory.Finally, two new algorithms and mentioned several filtering methods are compared through MATLAB simulation. The signal-to-noise ratio curve shows SNR has increased significantly after the new threshold algorithm. Therefore, the new threshold algorithm for denoising processing would get a clearer voice signal, and get to achieve the desired effect.