Multi-modal Speaker Authentication System Based On Speech And Lip Motion Correlation

Information security has become the focus of attention today,a single password verification has long been eliminated by the times,the main means of authentication is "what you know" and "what you have".such as mobile phone verification code,email authentication and other auxiliary verification methods.And many of the biological characteristics of human beings from the "who you are" principle,with a unique and measurable,and has easy to collect,authentication hidden,good recognition and high cost of the characteristics of high cost.Therefore,Biological information can be applied as a new authentication model.The remote speaker authentication system developed in this laboratory is based on the Chinese elderly social security fund to receive the identity function for the needs of the background of the project,with voice-print authentication,anti-recording playback attacks and other major functions,this article will be in the original speaker certification System to increase the voice lip to determine the same,prosthesis detection,face recognition and other functions,to further enhance the accuracy and robustness of the system to ensure that the user's information security.The specific work of this paper is as follows:(1)The speech module of the speaker authentication system is improved,the voice lip consistency detection module,the face authentication module and the multi-model fusion authentication module are added to the system frame,and the interface between the modules in the system is re-adjusted.So that the system can run and implement the authentication function.(2)Speech and lip motion correlation algorithm,the audio and video data through the preprocessing module,feature extraction module,scoring decision module processing,each module includes image and audio two lines,according to the different types of different interfaces.And finally get the confidence score,through the comparison threshold to complete the voice lip consistency judgment.In addition,on the basis of the original algorithm,some improvements have been made to make it more suitable for the current speaker system.(3)Face certification method design.Our system is based on the Fisher criterion,theLDA algorithm is used to classify the data,and the Euclidean distance between the unknown sample and the existing sample is used as the basis of judgment.Based on the face authentication,this paper attempts to integrate the method of face fusion with the face recognition.The experimental results show that the new fusion method can improve the accuracy of the original method to a certain extent.(4)Multi-modal fusion authentication method,the use of vocal print authentication module and the face recognition module to further improve the accuracy of the parameters of the parameters.In this paper,adaptive weighted fusion method is used to influence the aspect of system authentication decision.At the same time to a certain extent,to overcome the vocal print and face recognition of the inherent defects.