Research On Robust Voiceprint Identification Technology In Noisy Environment

With the development and popularization of the Internet and the emergence of information security problems,identity recognition technology has become an important development direction and difficulties in human-computer interaction and information exchange.Voiceprint identification,as one of the means of biometric identification technology,has been accepted and used gradually by people because of its convenience,visualization and so on,and has become a research hotspot in the category of biometrics,and has been applied in many fields.As a branch of speech recognition,the essence of voiceprint identification is to extract information that reflects personal characteristics from the speaker's pronunciation signal,and through the classification and matching of these personality information to complete the identification and confirmation of the speaker.In recent years,with the continuous development of voiceprint identification technology,the emergence of Gaussian Mixture Model-Universal Background Model,Identity-Vector,Probabilistic Linear Discriminant Analysis and other technical methods has created a new situation of voiceprint identification,so that voiceprint identification is gradually moving out of the laboratory towards practical application.However,in the real application environment,there will inevitably be a variety of background noise affecting the quality of speech,so that the actual application of the recognition system effect can not meet the ideal requirements,and noise becomes an important factor restricting the development of speech related technology.Therefore,this paper has carried on the thorough research to the robust voiceprint identification technology under the noise environment,and fully studies the problem of anti-noise in speech feature extraction and recognition model,which has the important theory significance and the broad practical value.The specific content of this article includes the following aspects:First of all,this paper summarizes the related technologies of voiceprint identification,and expounds its principle in detail.Then,in the aspect of speech feature extraction,a robust speech feature extraction algorithm,NPGFCC(Nonliner Power-function Gammachirp Frequency Cepstral Coefficients),is proposed by combining with Gammachirp filter which accords with the characteristics of human hearing.This feature is based on compression normalized Gammachirp filter bank,can accurately characterize the speech signal,and through the use of segmented nonlinear power function transformation,time series filtering and other ways to enhance its anti-noise performance,then the performance of the algorithm is analyzed by experiments in combination with the PLDA i-vector model.Secondly,in the aspect of voiceprint identification model,based on the mismatch between model training environment and real application noisy environment,this paper fully studies the Gaussian Mixture Model-Universal Background Model from the perspective of compensation.Combining with the idea of Parallel Model Combination,a feature compensation algorithm for adaptive noise estimation APMC is proposed.APMC feature compensation algorithm,which is robust to noise,can effectively reduce the mismatch between training environment and testing environment so as to improve the recognition accuracy and anti-noise performance.The experimental results indicate that the proposed method can reconstruct the clean speech GMM parameters more accurately.Also,this method can significantly improve the speaker identification accuracy,especially in low signal-to-noise ratio(SNR).Finally,this paper studied the popular voiceprint identification model of Identity-Vector and PLDA.By combining the algorithm of robust speech feature extraction and adaptive feature compensation and the idea of decision fusion.Based on the i-vector and PLDA models,a set of anti-noise voiceprint identification algorithm is constructed.The experimental verification and system simulation of the proposed method are carried out.The results show that the proposed method can effectively improve the anti-noise robustness of the system.