The Research Of Speech Denoising Algorithm Based On Wavelet Threshold And The Modulus Maximum Value

In the real life, the speech signal will be affected inevitably by the noise interferencewhich usually comes from the surrounding environments, the transmission medium and soon in the speech signal transmission process. Therefore, Speech denoising is a verynecessary means for speech signal processing. After many years of researches, Peoplehave put forward a variety of denoising algorithms about Speech denoising, such as theWiener filter method, the spectral subtraction, and the harmonic enhancement method.Wavelet transform is a new mathematical tool of analysis, as it is known, it is a multi-scalesignal analysis method which uses the telescopic factor and translation factor onmulti-scale signal analysis. At present, the speech denoising has been widely used in ourlives. In this paper, basising on the wavelet transform, it will carry out the followingresearches around the threshold denoising method and the modulus maximum valuemethod.Firstly, the paper will enlaborate the method basing on the wavelet threshold denoising,and it will specially focuse on wavelet decomposition level, wavelet basis, and selectionrules of threshold parameters in the process of wavelet denoising. Basing on the traditionof the hard and soft thresholding function denoising, this paper will research an improvedthreshold function by combining the advantages and disadvantages of the two methods. Itcan be seen from the Matllab simulation results comparison of MSE and PSNR that theimproved method has better denoising effect.Secondly, the wavelet modulus maximum denoising method was discussed, as it isknown, the modulus maximum value method is a classical denoising methods. Becausethe wavelet transform modulus maxima as the scale increases, while the noise of thewavelet transform modulus maxima average density will reduce as the scale increases, andthe characteristics will be used as the basis of denoising method. Basing on modulusmaximum denoising, this paper will research an improved modulus maxima methodwhich combines an adaptive threshold and the modulus maximum value method together.Simulation comparison showes that the algorithm has good noise reduction effect. Finally, in order to further analysis the denoising effect by wavelet threshold methodand the modulus maximum value method, the paper will compare the denoising effect indifferent input SNR, and obtain that the wavelet modulus maxima method has betterdenoising effect in low SNR than the threshold method.