Speech de-noising method based on empirical mode decomposition and improved wavelet threshold

To restrain the noise introduced during the transmission of speech signal effectively,a method based on EMD and wavelet threshold de-noising was proposed.Moreover,aiming at the problems of the discontinuance of hard threshold function and the constant deviation between estimated wavelet coefficients and decomposition wavelet coefficients in the soft threshold function in conventional wavelet threshold de-noising,a new kind of high order derivable threshold function was constructed,which could change the shape of the function flexiblely by adjusting dual parameter to get close to the ideal wavelet coefficients.The EMD-based wavelet threshold method was applied to process actual speech signal.Simulation results showed that the proposed method increased output SNR and restrained the noise better,compared with speech de-noising based on wavelet and EMD scale filter in the case of low SNR,and could be used in the front of speech recognition system in noisy environment to improve the accuracy of the recognition results.