Based On Wavelet Transform And Emd Denoising Containing Noise Aliasing Blind Speech Separation

While the rapid development of society, the rhythm of society is more and more fast. The language as the most direct and quickest communication method of human, we need to pay more attention to that how to deal with the voice signal better.At present, many algorithms of the signal processing have achieved relatively good results, but most of the algorithms only deal with the single which just has pure voice, the consequent of processing the mixture voice signal which containing noise is terrible. The purpose of this thesis is take a pre-processing of the noisy mixture voice signal, in order to lay a good foundation for such as voice recognition system for application.Now, the noisy voice signal processing have two methods, there are generally divided into auditory scene analysis and blind source separation. The blind source separation method has been used widely because the method of it is very simple the theoretical of it is very maturity and the algorithm of it is very effective. However, the noisy voice separation results will decrease terribly. Therefore, it is necessary to conduct a de-noising voice processing first.The traditional method of removing noise from the signal are generally based on the Fourier transform, while the Fourier transform approach is more suited to deal with the stable signal, the effect of non-stationary signal processing don't have the result which desired to be. Voice signal as the typical non-stationary signal, commonly we used wavelet transform method for processing.However, Wavelet transform usually has limitations in the dealing with noise. EMD is a new method of the non-stationary signal processing which proposed by Huang in the late 20th century, it can decompose the signal into several IMF components which have progressively decreasing frequency, because most noise components are concentrated in the high-frequency, so it should be better than process the signal only by wavelet transform, if we process the high-frequency IMF which decomposed from the signal by EMD. Therefore, this paper will joint application of wavelet transform and EMD to deal with the noising voice signal.In this paper, proposed a new de-noising method in the mixture noising audio signal, which joint the wavelet and EMD to a secondary de-noising. Through the selections of the threshold selection rules, decomposition levels, wavelet function, and which IMF will process secondary by using simulation. Verify the feasibility of this method, and discussed the influence by the selection of details. This kind of audio processing method make a good result in the mixture noising audio signals, it has a broad prospect in application.