Speaker Diarization Using EHMM and CLR

To solve the problem of speaker information deficiency affecting accuracy of segmentation in the conventional speaker diarization system,this paper proposes the method of multistage speaker segmentation and clustering algorithm based on EHMM and CLR.On the basis of traditional algorithm,we develop a resegmentation and reclustering scheme which combine with unsupervised segmentation algorithm based on GLR distance measure and a hierachical clustering algorithm based on BIC.Test on the national institute of standards and technology(NIST) RT′03S corpus,by comparing the conventional one,more than 41% of relative diarization error is reduced.