Deep Learning Based Speaker Authentication With Random Password

Biometric-based identity authentication systems have attracted more and more researchers' attention,and biometrics such as faces and fingerprints are widely used in many access control systems.In recent years,some studies have shown that the human lips area and the movement of the lips during speech contain a large amount of identity-related information,and lip-based identity authentication has received a lot of attention.Moreover,the lip-based identity authentication system can be well combined with other biometrics,and has great potential in user acceptance and industry landing.Therefore,it is important to study lip-authentic identity authentication.Good certification performance and liveness detection are two key requirements of many certification systems.In order to avoid replay attacks,this paper proposes a new lip-based speaker authentication scheme with random passwords.Speaker authentication for random passwords is more challenging than fixed password scenarios because it is not possible to ask the system user to speak all possible passwords as training samples.In order to solve this problem,this paper proposes a new deep convolutional neural network,which consists of three sub-networks,namely lip feature extraction network,identity authentication network and content recognition network.In the lip feature extraction network,a series of spatio-temporal residual units are used to fully describe the static and dynamic characteristics of the lip biometrics.Combined with the characteristics of identity authentication task and content authentication task,this paper designs the identity authentication network and content recognition network,and proposes an end-to-end multi-task learning scheme,which can optimize the weight of the above three sub-networks simultaneously.This paper evaluates the authentication performance of both fixed and random passwords through experiments.The experimental results show that compared with the existing methods,the proposed method can obtain superior performance under the condition of fixed password.In addition,in the random password scenario,the method also obtains satisfactory verification results,which provides a reliable solution for user identity verification and also ensures the function of live detection.