Study On The Voiceprint Recognition Technology Based On Speech Separation

Now with the rapid development of information technology and the improvement of science and technology level,the national society need of the information security is higher and higher,and biometrics combines biology and informatics,the identification is more convenient and secure.Among all the biometrics technologies,because of the low cost of voiceprint,it gets relatively simple processing advantages,an important research direction from the beginning of the field of biological identification.In recent decades,the field of voiceprint recognition has gained great progress,which is widely used in national defense,finance,criminal investigation,medical and other civilian and military security.But in multi speakers,or strong noise background,voiceprint recognition effect is still not ideal.Aiming to solve the low efficiency of the recognition under multi speakers' voice background,we need to separate the voice before recogniting.This thesis focuses on the separation and extraction of speech feature,the specific description of voiceprint recognition and other related algorithms,the voiceprint recognition system based on speech separation research.This paper begins with an overview of the basic theory of speech separation and voice recognition,and then the fusion feature extraction,speech separation,model training and pattern matching launched a detailed description.In the feature extraction stage,through the integration of pitch and Mel frequency cepstrum parameters and poor fusion parameters of PITCHMFCC parameters to generate 3L+1 dimension.In the phase of speech separation,the speech scene is analyzed by auditory scene analysis.In the stage of voiceprint recognition,voiceprint recognition analysis classic classic models GMM,CASA integration of voice separation and recognition of SGMM model,and compare the two models of voiceprint recognition in both single and mixed speech scene recognition accuracy,in addition,also analyzes the different gender mixed speech,the effects of different distance between different background speech length and background audio speaker when the phone on the SGMM model and the recognition rate.Finally,the voiceprint recognition system based on speech separation,and the system module and process for detailed description and display.The main work of the paper is as follow: Propose the fusion feature parameter based on pitch and MFCC.Analysis and simulate the CASA speech separation model and GMM voiceprint recognition model.Propose the SGMM model which fuse the speech separation model and voiceprint recognition model and prove the validity of the model by experiment.Finally make the voiceprint recognition system based on speech separation with programming.