Application of Wavelet Neural Network in Speech Signal Denoising

In order to solve the interfering problem in the process of speech communication, we aimed at the property of speech and noise, and then improved wavelet threshold denoising method based on multi-resolution analysis. By combining with neural network, a self-study threshold wavelet denoising method was presented We used soft, hard threshold and self-study threshold methods to denoise a speech signal respectively. Experimental results demonstrate that self-study threshold wavelet denoising method is more efficient to enhance speech signal. This method can amend a residual musical noise in traditional methods. It can also greatly improve the SNR and definition. It mainly has some advantages as the following: (1) It does not need the statistical characteristic. (2) These thresholds are self-adjusting according the signal.