Research On The Extraction Method Of Speech Feature Parameters In Voiceprint Recognition

Voiceprint recognition,also known as speaker recognition,is to identify the speaker's personal identity using the speaker's voice information.In recent years,with the rapid development of Internet,voice print recognition technology has been paid more and more attention.As a kind of biometric authentication technology,it has many advantages.It has been applied in the fields of bank identity authentication,remote attendance system of the company,and close voice human-computer interaction.Voiceprint recognition system is mainly divided into two parts: feature parameter extraction and recognition model establishment.The quality of feature parameter extraction is directly related to the recognition result,so feature parameter extraction is the most important part of the voice print recognition system.The speaker's speech contains personality information and common information.Personality information is caused by differences in individual pronunciation organs,pronunciation channels and pronunciation rhythm.Common information is the same basic unit in the same language composition.The purpose of feature parameter extraction in voiceprint recognition is to extract the personality information.However,the common information and personality information are intertwined,so it is difficult to extract only the personality information at present.Often,the extracted information contains the common information.Therefore,this paper focuses on the extraction method of speech feature parameters in striation recognition.1)Based on linear prediction coefficient(LPC)and buck cepstrum coefficient(BFCC),blend in LPC BFCC,by linear prediction cepstrum coefficient(LPBFCC)buck,because the human voice of the voiceless and voiced energy difference is very big,and on the basis of LPBFCC join each frame of short-time energy,combination of characteristic parameters(ELPBFCC),used for voiceprint recognition.In the pure environment,compared with MFCC and BFCC,the recognition rate increased by 2.62% and 2.46% respectively.In addition,in the noise environment,the recognition effect is not ideal.LPC is sensitive to noise based on the characteristics of the channel model,and the noise details are also incorporated into the BFCC,resulting in poor robustness.2)Triangular filter banks are used to extract the common feature parameters.Their internal downward trend of a single filter is fast and smooth,which leads to loss of connection with the front and back sub-bands,thus affecting the accuracy.In this paper,Gaussian shaped filtes are used instead of triangular filter Banks,which to some extent make up for the deficiency of triangular filter Banks.The internal descending trend of filter Banks is slowed down,the smoothness is increased,and the contact between adjacent subbands is enhanced.3)In view of the insufficient high-frequency resolution of the barker frequency cepstrum coefficient(BFCC)with Gaussian shaped filtes,this paper proposes the inverted barker frequency cepstrum coefficient(IBFCC)with Gaussian shaped filtes.Then,combining the advantages of high resolution of BFCC in the low frequency part and high resolution of IBFCC in the high frequency part,Fisher criterion was proposed to screen and combine BFCC and IBFCC into a new mixed feature parameter(FBAI).4)The pattern recognizer Gaussian mixture model(GMM)is used as the recognition model to test the recognition performance of the above characteristic parameters in the system.Simulation results show that the recognition rate of FBAI is 3.32%,3.23% and 0.91% higher than that of MFCC,BFCC and ELPBFCC respectively in the environment of noiseless speech.Under different SNR in the noise environment,the recognition rate of FBAI is improved compared with that of MFCC,BFCC and ELPBFCC.On the whole,with the increase of SNR,the improvement of recognition results gradually decreases.To sum up,FBAI parameter can effectively improve the performance of voice print recognition system and has better robustness.