Based On The Wavelet Transform Of Speech Denoising Research And Application

The speech signal is a time-varying, non-stationary random signal. People in the process of voice communication will be from the surrounding environment, transmission medium and other factors caused by noise. When noise frequency spectrum is similar to that of a voice, a classic traditional voice signal de-noising methods such as time domain and frequency domain, add window Fourier transform, adaptive filtering, wiener filtering, etc. Each has its limitations, for the white noise is often difficult to achieve the good de-noising effect. limits the range of application. The wavelet transform can be good for time-frequency analysis of the signal to overcome the shortcomings of the classical de-noising method.Wavelet transform is based on the continuous development of short time Fourier transform(STFT), but also overcome the window size does not vary with the frequency change and other shortcomings, can be very good to the signal time-frequency analysis and processing. At the same time, it is also a multi-scale signal analysis method which uses the telescopic factor and translation factor on multi-scale signal analysis. In this paper, the wavelet threshold method of noisy speech signal de-noising analysis, threshold method in the four important parameters include: wavelet base, wavelet decomposition layers, threshold function and threshold for a detailed description and simulation. So by comprehensive method one: frame for noise speech signal, then using the simulation with four important threshold parameter values, for different frame the SNR signal corresponding wavelet threshold de-noising processing; and another integrated approach: using the characteristics of speech signal is composed of voiceless and voiced, and combined with wavelet threshold de-noising to deal with the noise was carried out on the noise speech signal processing.Through four basic simulation experiment get wavelet voice threshold de-noising of four important parameters: the wavelet base for sym8 wavelet, wavelet decomposition layers of 3 layers, the threshold function for the soft threshold function, the threshold for heuristic threshold. Also a synthesis method is obtained by simulation experiment and comprehensive method 2 is composed of four important parameters in front of the general wavelet speech de-noising method has better de-noising effect.