The Research Of Speaker Recognition Method Based On Wavelet Transform

Speaker recognition systems can obtain very high accuracy for clean speech, but their performance will degrade rapidly in noisy environments owing to the mismatch between the acoustic models and the testing speech. Therefore, noise robust technology is acrucial problem for the application of speaker recogcition system in real life.This paper focus on the research of the speech denoising based on wavelet transform .we introduce the basic theory of wavelet methods of signal denoising based wavelet transform and application of de-nosing methods in speech. In these algorithms hard threshold methods have limitations and are not effective. Now we present a new method .This method can supply a gap. Is a good hard and soft threshold to improve the program, used to overcome the hard-threshold noise ineffective and soft threshold so that the signal distortion smooth over shortcomings. When the difference of the wavelet coefficients of signal and noise in critical point is great, we can choose threshold easily. Based on the above considerations this paper, a mean approaching the threshold down throat with a combination of methods, which use an improved thresholding function, and finally through the experimental verification of the algorithm effectiveness and superiority.