说话人识别中测试时长与识别率关系研究  

Investigation of the Relationship between the Length of Testing Speech and the Accuracy of Speaker Recognition

作　　者：孙林慧[1] 叶蕾[1] 杨震[1] 

机构地区：[1]南京邮电学院信息工程系,江苏南京210003

出　　处：《计算机仿真》2005年第5期231-233,258,共4页Computer Simulation

基　　金：江苏省'青蓝工程'基金(QL003YZ)。

摘　　要：测试时长是影响说话人识别问题的主要因素之一。该文主要对分布式语音识别中测试时长与说话人识别率的关系进行了研究。文中采用文本无关的训练模板,首先对基本的说话人辨认系统用干净语音和带噪语音进行了测试,结果表明系统识别率随测试时长的增加而提高,并在实验室条件下获得加噪语音最佳测试时长。其次为了减小最佳测试时长采用改进的说话人辨认系统,先对说话人的性别进行分类然后再对其身份进行识别,不仅减少了测试所需的最佳时长,而且提高了系统的抗噪性能。最后对仿真结果进行了分析。<Abstrcat>The length of testing speech is one of the key factors of affecting the speaker recognition. This paper focuses on the relationship between the length of testing speech and text independent speaker recognition accuracy of the system for DSR. The first experiment, which is done with clean and noisy speeches, shows that the longer the testing time of input speech signal, the higher the average recognition accuracy we get, and that we can obtain the optimal testing length for noisy speeches . In the second experiment, we first of all classify speakers into two groups by speakers' gender, and then we recognize their identities in some groups,which reduce the optimal testing length for noisy speech and enhance the robust of system . Finally, the experiment results are discussed.

关 键 词：分布式语音识别 测试时长 说话人辨认 识别率 

分 类 号：TN912.3[电子电信—通信与信息系统]