Abstract:The traditional wavelet threshold denoising method can cause the loss of useful speech signal, and the improvement of signal to noise ratio is not ideal. By analyzing the principle of wavelet denoising, an improved speech enhancement method of wavelet threshold function is proposed. The method combining the advantages of wavelet denoising for soft and hard threshold function, which can overcome disadvantage of discontinuous and bias for hard threshold function and soft threshold function respectively. The method makes judgment for voiced signal in the first place, then uses changed threshold to the wavelet coefficients of voiced and unvoiced signal in processing. The simulation experimental results show that the improved method is very suitable for speech enhancement under strong noise background, no matter keeping the voiced speech information in the noise speech signal or improving signal-to-noise ratio index, the improved method is better than traditional soft threshold method, the spectrum subtraction and the wavelet transform of auditory sensation.