Abstract：Aiming at the problem that the speech signal suffers great attenuation by a Wiener filter under complex background noise, a speech enhancement method is proposed that combines multi-window spectrum estimation and a Wiener filter. First, In this method, a multi window spectral estimation is performed with a noise speech, then, approximate pure speech spectrum is obtained by wavelet threshold techniques, the obtained approximate pure speech spectrum is compared with the speech spectrum that has been processed by the Wiener filter, and the proper spectrum is obtained as the final enhanced speech spectrum according to the distortion type. The simulation results show that under different types of noise and signal-to-noise ratios, the method is superior to the mean square cross prediction error ( MSCEP) method and prewhitening subspace ( PSS) method in suppressing noise and reducing speech signal attenuation situation.

Keyword：SNR;wiener filter;wavelet thresholding;multi-window spectrum estimator;