The Study Of Speaker Identification Based On The Relavant Vector Machines

Speaker recognition is the branch of the important application of the artificial intelligence.And speaker recognition has been widely used in many fields, such as the judicial judgment, the E-banking services, the security monitoring and the entertainment, because of its convenience, noncontact and economy. The speaker recognition technology currently has gradually moved toward the applied part from the theoretical part. The evaluation standard of speaker recognition system is not only the recognition rate, but also the enhancement of the efficiency of model construction, the system operational speed, the extension of algorithm and the recognition rate.The thesis is mainly about the design and the implementation of speaker recognition system by usage of the better kernel function classification tool against a background of limited samples.By adding the training algorithm, it has achieved the rapid model contruction, avoided the inverse operation of the large matrix and also sinplied the calculation amount of the training process. At the same time, the speaker identification task based on frame can be realized, and the algorithm running efficiency can be improved greatly.According to the applying requirement of the real biometric identification, it can design and implement the identification lock with higher accuracy and more real-time by practically using RVM and Multivariate Expansion Model in Vioce Print Lock.The author applies the RVM to Vioce Print Lock in this thesis, and develops with programs. The application system is including the language input module, Speech signal preprocessing module, speech feature extraction module, speaker recognition module, Database management and interface module. The usage of this algorithm can improve the accuracy and real-time of the system partly.