The Research On Speaker Recognition In Noisy Environment

Prevailing speaker recognition systems can obtain very high accuracy for clean speech,but their performance will degrade rapidly in noisy environments owing to their mismatch between the acoustic models and the testing speech.Therefore,noise robust technology is a crucial problem for the application of speaker recognition system in real life.The paper focus on two key problems of speaker recognition system,which contains the following aspects:Firstly,this paper mainly studied MFCC's extraction theory and the application of wavelet analysis to speech signal processing.What's more,a method was proposed for extraction of a new feature parameter FPBW,which sub-band energy's calculation and sub-band energy's separation both based on wavelet analysis.The experiments showed that the new parameter FPBW had better recognition rate and noise-resistive property than traditional parameter MFCC.Secondly,owing to speech signal contaminated by white noise;firstly,Multi-scale wavelet analysis was applied to the process of signal preprocess,based on it,the theory of multi-scale analysis was applied to separate speech and noise,and enhanced the speech consequently.Secondly,de-noised speech input into original speaker system. The experiments showed that this method eliminated noise effectively,and improved speaker recognition rate in noisy environment.Thirdly,a new robust speaker recognition method was proposed which combined wavelet de-noising technique with the new parameter FPBW.The experiment showed that it effectively improved speaker recognition in noisy environment.Lastly,the Gaussian mixture model(GMM)techniques were increasingly used in speaker recognition.Aimed at the problem that wavelet transform needed large computation and GMM needed long training time,an orthogonal guass mixture model (OGMM)was employed in order that orthogonal transform was performed before the use of expectation Maximization algorithm,so that the computational speed was improved.