说话人识别中基于最大后验概率的通道补偿方法  

Channel compensation of speaker identification based on maximum a posteriori

作　　者：高荣春[1] 韩纪庆[1] 张磊[1] 

机构地区：[1]哈尔滨工业大学计算机学院,黑龙江哈尔滨150001

出　　处：《通信学报》2009年第3期99-103,共5页Journal on Communications

基　　金：国家重点基础研究发展计划("973"计划)基金资助项目(2007CB311100);国家自然科学基金资助项目(60672163)~~

摘　　要：为了解决通道变化对说话人识别系统性能的影响,将最大后验概率方法应用到具体的通道补偿中。首先设置标准通道,然后利用其他通道和标准通道偏差的分布作为先验知识,通过少量自适应数据获得当前通道的偏差对特征进行调整,达到补偿的目的。实验结果表明,在1s自适应数据时,该方法的识别率达到89.1%,和倒谱平均减方法相比识别率提高4.2%,随着自适应数据的增加,采用该方法的系统性能可进一步提高。To solve the effect of channel changes on the performance of speaker identification system, apply the method of maximum a posteriori to specific channel compensation. First, setup a standard channel, then, use the distribution of deviation of other channel and standard channel as experience knowledge, and obtain the deviation of current channel by a few adaptive data, use these deviations to adjust the characteristic to gain the goal of compensation. The results show that adaptive data in 1 s, it has a discrimination of 89.1%, above the method of cepstrum average reduction by 4.2%,with the increasing of self-adaptive, the performance of this system can be improved.

关 键 词：说话人识别 通道补偿 最大后验概率方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]