摘要
提出一种新的说话人验证方法。本方法通过对用户登录的语音数字串的内容进行识别来验证用户身份。而且采取计算后验相似度比例对数LLR(log-likelihood ratio)的措施,进一步提高系统的可靠性和识别准确率。提出的方法与目前普遍使用的说话人验证方法相比,本方法所需要的训练语音相对较少,对环境的要求也比较低,因而注册使用更加方便,保密性更高。通过实验,系统的合法用户通过率达到98.4%,非法用户拒识率达到100%。
A new method for speaker authentication is presented, which recognizes user authentication by recognising the serials of number speaked by user and compareing it with the user registed data. In addition, the reliability and veracity by calculating the log-likelihood ratio are increased. Compared with common method for recognizing user authentication, this method needs less training, and more convenience for use, and can improve the reliability further. The result of simulation shows the pass rate of the legality user is over 98.4% and the reject rate of the falsity user is 100%.
出处
《测控技术》
CSCD
2005年第9期7-8,14,共3页
Measurement & Control Technology
基金
国家自然科学基金资助项目(60272040)
关键词
内容识别
说话人验证
相似度比例对数
content verification
speaker authentication
log-likelihood ratio
