New threshold function of wavelet packet speech enhancement algorithm
一种新阈值函数的小波包语音增强算法

Aiming at enhanced speech by using the traditional soft and hard threshold functions methods had the problem of distortion, this paper proposed a new threshold function of wavelet packet speech enhancement algorithm, and put forward the new threshold function and the new Bark scale wavelet packet decomposition structure. The new threshold function flexibly combined the soft and hard threshold functions when wavelet packet coefficients were greater than the given threshold. When wavelet packet coefficients were less than the given threshold, it used a nonlinear function to instead of setting wavelet packet coefficients to zero simply in the traditional threshold function, and it achieved the smooth transition of the threshold function. The new 60 sub-bands Bark scale wavelet packet decomposition structure could achieve better simulation of the human ear auditory perception characteristics. Simulation experimental results show that comparing with traditional threshold function methods, the new proposed algorithm improves the SNR and SSNR of the enhanced speech significantly and reduces the speech distortion in white Gaussian noise and colored noise background, and has better denoising effect.