说话人聚类的初始类生成方法  

Generating initial clusters for speaker clustering

作　　者：赖松轩 李艳雄[1] LAI Songxuan;LI Yanxiong(School of Electronic and Information Engineering, South China University of Technology, Guangzhou 510640, China)

机构地区：[1]华南理工大学电子与信息学院,广州510640

出　　处：《计算机工程与应用》2017年第3期149-153,共5页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61101160);广州市珠江科技新星专项(No.2013J2200070);华南理工大学大学生课外创新训练项目;中央高校基本科研业务费本科生自主选题项目(No.10561201501)

摘　　要：目前说话人聚类时将说话人分割后的语音段作为初始类,直接对这些数量庞大语音段进行聚类的计算量非常大。为了降低说话人聚类时的计算量,提出一种面向说话人聚类的初始类生成方法。提取说话人分割后语音段的特征参数及特征参数的质心,结合层次聚类法和贝叶斯信息准则,对语音段进行具有宽松停止准则的"预聚类",生成初始类。与直接对说话人分割后的语音段进行聚类的方法相比,该方法能在保持原有聚类性能的情况下,减少40.04%的计算时间;在允许聚类性能略有下降的情形下,减少60.03%以上的计算时间。During the procedure of state-of-art speaker clustering,the individual speech segment directly obtained fromspeaker segmentation is used as an initial cluster,which leads to huge amount of calculation.In this paper,an algorithm ofgenerating initial clusters for speaker clustering is thus proposed in order to reduce calculation load.First,features areextracted from speech segments,and centroids of features are calculated.Then the initial clusters are generated by clusteringthese centroids using both hierarchical clustering algorithm and Bayesian information criterion under an easy stoppingcriterion.Experiments show that doing speaker clustering on the initial clusters generated by the proposed method is fasterthan doing speaker clustering on the speech segments directly obtained by speaker segmentation.The computationalreduction is about40.04%without losing the performance of speaker cluster,and the computational reduction is morethan60.03%with losing little performance of speaker cluster.

关 键 词：层次聚类 贝叶斯信息准则 说话人聚类 初始类 语音信号处理 

分 类 号：TN912.3[电子电信—通信与信息系统]