The Algorithm Research Of Feature Parameters Extraction For The Speaker Recognition Based On Wavelet Packet Transform

With the rapid development of Internet technology,the speech features have become an important resource for the computer processing.Information Processors like traditional texts,images and other information processors have huge market in some areas.However,in some areas information processors like texts,images can't reflect the characteristics of the specific unit,and they also have limitations on the certification of the specific unit.Speaker recognition is a kind of authentication technology by extracting and analyzing human voice signal and then judge who is the speaker.Speaker recognition technology in some areas can achieve the certification of the specific unit through analyzing the differences between different signals to achieve the goal to tell the differences from people.With the development of technology,this technology will be applied to telecommunications,Internet,health care,electronics and other various fields.Therefore,the study of speaker recognition technology is of great significance.The thesis is based on the basic theory for the characteristic parameters of MFCC feature extraction process,the principle of wavelet packet transform and Teager energy operator,and puts emphasis on the speaker recognition process of wavelet packet transform and conducts research on speaker recognition process based on wavelet packet transform and energy operator.The main work is summarized as follows:First,the thesis will make a Brief introduction of the speaker recognition technology.Meanwhile it makes an elaborate introduction of several commonly used speaker recognition technology,and the advantages and disadvantages of the traditional calculation method based on MFCC were analysised.In addition,the principle and theory of wavelet packet transform were analysised and introduced.A feature extraction algorithm based on wavelet packet transform is proposed.The thesis analyzed the advantages and disadvantages for MFCC.The wavelet packet transform was applied to MFCC extraction process,using the wavelet packet transform instead of Fast Fourier Transform and Mel filter banks,which can reflect the characteristics of signals : non-stationary,shortly sustained,Time domain localized and Frequency domain localized,and at the same time solve the problem of accuracy which is not high in Mel filter frequency,finally we got the new characteristic parameters called newMFCC.In addition,we also combined newMFCC with the spectral centroid and extracted their combination feature parameter.Finally we use the differential parameters as the characteristic parameter for comparison.The experimental results show that the proposed algorithm can achieve high recognition performance.A wavelet packet transform and Teager energy operator are combined as a new feature extraction algorithm.The thesis analyzed the advantages and disadvantages for MFCC in the noise environment,and combined with the performance of Teager energy operator in the enhancement of speech.Then Teager energy operator is applied to the previous feature extraction algorithm based on wavelet packed transform.The process of feature extraction algorithm was constructed and the new characteristic parameters named newMFCC2 from the experiment is raised.Finally,the algorithm proves better recognition results in noise environment.