Telephone Conversation Speaker Recogniton System Based on Speech Purify

In this paper, we present a telephone conversation speaker verification system. We use the most popular approaches which based on distance measure and then followed by a hierarchical clustering to segment the speaker from the conversation. In order to get cleaner speech, the approach based on Purity Criterion is introduced, it uses hierarchical clustering first and then scores the segment result again, the higher score speech segmentes are reserved while the other are discarded. we carry out experiments on the NIST´2006 SRE evaluation data set. The results show that the introduced method achieved better performance in equal error rate (EER).