一种基于改进的矢量量化算法的说话人识别方法  

A Speaker Recognition Method Based on Modified VQ Algorithm

作　　者：张云雁[1] 张建军[1] 赵风光[1] 吴龙梅[1] 

机构地区：[1]上海大学理学院,上海200444

出　　处：《上海大学学报（自然科学版）》2005年第4期368-371,共4页Journal of Shanghai University:Natural Science Edition

基　　金：上海高校发展基金资助项目(214348)

摘　　要：该文针对LBG算法可能有空胞腔产生及有些码字利用率低的问题,提出了一种改进的矢量量化算法,并将其应用到与文本无关的说话人识别研究,得到了一种新的说话人识别方法.实验表明,这种方法对说话人的识别性能好于基于LBG算法的说话人识别方法.The traditional LBG algorithm is an important VQ method, but it has two disadvantages. First, empty cells may be produced, and secondly, some code vectors may be used only by very few training vectors in the LBG algorithm. In this paper, a modified LBG algorithm is proposed, and the design of a textindependent speaker recognition system based on the modified VQ algorithm is presented. Experiments show that the system has better performance than that based on the traditional LBG algorithm.

关 键 词：矢量量化 LBG算法 说话人识别 LPC倒谱系数 

分 类 号：TN912.34[电子电信—通信与信息系统]