A speaker identification system with verification method based on speaker relative threshold and HMM

This paper presents a speaker identification system with speaker verification method. Compared with usual speaker identification systems, it requires much less speech data but still maintains high performance. In the implementation of this speaker identification system, we introduced several approaches to improve the effect of speech segmentation and used a speaker verification method based on a common codebook, HMM and speaker relative threshold (SRTHD). A novel method of integrating the speaker identification with the face recognition to verify a person's identity is also presented. It has been proved that this method is efficient to reduce the equal error rate of recognition.