Research On Speech Enhencement Based On Wavelet Analysis

In recent years, with the rapid development of electronic technology and network technology, digital voice signal processing technology for its simplicity, efficiency and other advantages of wide range of applications in various fields. At present, speech coding, speech synthesis and voice recognition technology has matured and has been widely applied in real life. In real-world usage, the noise will reduce the performance of the voice signal processing system. Therefore, to enhance the speech signal in the preprocessing stage is of practical significance. Wavelet threshold denoising method based on the noise signal and voice signal in the wavelet domain of the distribution characteristics of each layer of the wavelet coefficients thresholding, can effectively eliminate the noise component, this article take the wavelet threshold denoising method with noise The voice signal for voice enhancement processing.Depth analysis of the wavelet analysis in digital signal processing applications, focusing on the application of multi-resolution wavelet decomposition of the signal noise reduction processing. Depth analysis of the threshold noise reduction algorithm of the threshold value functions and threshold threshold signal denoising, reference-based spectral subtraction speech enhancement algorithms based noise extraction of the door threshold to determine the algorithm. Been tested, compared to several wavelet threshold denoising threshold estimation criteria, and the door threshold to determine the algorithm to take up less computing resources and can achieve good speech enhancement effect. Traditional threshold denoising algorithm based on experience to determine the decomposition level of decomposition levels will be less cause denoising is not complete, and the decomposition level is too high can cause signal distortion. In this paper, the decomposition level of the threshold noise reduction algorithm in-depth study, an adaptive optimal decomposition level determination algorithm. The distribution characteristics of noise signals and voice signals in the wavelet domain, singular spectrum analysis through the layers of wavelet coefficients, compared to the distribution of the singular spectrum of different SNR with noise signals is proposed based on wavelet coefficients singular spectrum characteristics to determine the optimal decomposition level of the algorithm. Been tested, the algorithm can determine the optimal decomposition level with a noise signal to noise ratio adaptive effectively increase voice enhancement and to avoid unnecessary waste of hardware resources.