Voiceprint Recognition Technology Research Based On Gmm

Speaker recognition technology, also known as the voice print recognition, is basedon human biological characteristics to determine the identity of the person. Sound as themost natural means of communication, with its incomparable advantages was widelyapplied to identification.For speaker recognition modeling there are a variety of techniques, GaussianMixture Model with its simply, good performance and text-independent feature is oneof the most frequently used method of modeling. This thesis describes the Gaussianmodel, parameter estimation and recognition method. For speech frames in certain voiceframe will affect the system recognition rate in the recognition phase,we give a votingbased method.Using gaussian mixture model in the speaker recognition,when thespeaker’s number is large then there need amount of calculation.We combine the VQmethod with Gaussian mixture model,the models are divided into two main parts whichare Male and Female parts, and we use dynamic time algorithm to calculate the distancebetween each pitch,then reduce the recognition time.At present,most work of speaker recognition technology study is based on theGaussian mixture model. In order to obtain a higher recognition rate we choose bettersound charcteristic parameters of the speaker and recognition algorithms. Speakerrecognition elaborates the characteristics of speaker recognition technology, extraction,modeling and other sectors. Recently, most speaker recognition method are usingMFCC and based GMM model.Another feature parameter of voice speech, pitch, isadded in this paper against imitative of MFCC.Adding Dynamic MFCC coefficients tothe feature vector will make the feature vector becomes complex, to shorten the time ofspeaker recognition,we given a weighted MFCC based on their contribution to theidentification rate.Experimental section is in the last part of this thesis, verify that the characteristicparameters of the Gaussian mixture model order, weighted MFCC, the recognition rateand the experimental results analysis.The experiment results show that the MFCC havea better performance than LPCC.When MFCC combine with Dynamic MFCC,the recongnition rate was obviously increased.The Weighted MFCC raises the recognitionrate and at the same time reduce the complexity of the calculation.We analysis pitch’sfunction and its effect on recongnition rate at last.