Abstract：Speaker clustering addresses the problem of grouping a set of speech utterances based on the identity of the speaker of the utterances.In this paper we proposed a novel clustering algorithm based on two distance metrics combining Generalized Likelihood Ratio and Normalized Cross Likelihood Ratio.In our proposal,Mel Frequency Cepstrum Coefficientsare first extracted from speech samples and modeled by Gaussian Mixture Models to represent the speech.Following a hierarchical clustering scheme is built combining GLR and NCLR metrics.In addition,Bayes Information Criteriais employed as the termination criterion.Experimental results show the cluster performance of combining GLR and NCLR is improved compared with either of them.As well,the efficiency is also improved greatly compared with the traditional GMM cluster method.
陈玥同,刘学亮. 结合两种距离测度的说话人聚类算法[J]. 小型微型计算机系统, 2015, 36(10): 2369-2373.
CHEN Yue-tong,LIU Xue-liang. Speaker Clustering Algorithm Based on Two Distance Metrics. Journal of Chinese Computer Systems, 2015, 36(10): 2369-2373.