The Speaker Recognition In Noisy Environment

It is an important problem that Automatic Speaker Recognition is applied in noisy environment. For speech signal is distorted by noisy signal, the trained speech model is not matched with the recognized one and the speaker recognition rate of system is severely affected.The paper focus on two key problems of speaker recognition system, which contains the following aspects:Firstly, the method of feature extraction is improved in a sort of way, that is PAC (Phase Autocorrelation) feature. In the paper, modified PAC feature is combined with Sigmoid function, and RASTA (Relative Spectral) Filtering is also applied to the speaker recognition. The experimental result shows that it effectively improve speaker recognition in noisy environment.Secondly, because of its structure and trait, sub-band process is widely used in the speaker recognition in noisy environment, so PAC feature and sub-band feature are combined, then recognition output is gained by probabilistic combination.Finally, in classifier design, the HMM (Hidden Markov Model) and WNN (Wavelet Neural Network) are joined according to their strongpoint. The experimental result shows that the method is not only availably classified, but also is of noise-resistive property.