一种基于遗传算法的话者身份确认系统建模方法  

SPEAKER RECOGNITION TEMPLATE BUILDING USING GENETIC ALGORITHM

作　　者：司虎[1] 戴蓓倩[1] 李辉[1] 

机构地区：[1]中国科技大学电子科学与技术系,合肥230026

出　　处：《小型微型计算机系统》2000年第1期79-82,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金

摘　　要：本文描述了一种采用短语音和基于遗传算法的话者身份确认系统的建模方法．利用遗传算法的全局搜索和优化特性，系统只需利用话者的短语音，就可快速建立话者的一类较优秀的模板．实验结果表明，这种方法既降低了用户的语音数据采集量，有利于话者模板的建立，又提高了系统的确认性能及鲁棒性，较传统方法有明显的优越性．In this paper, we describe an efficient template building method based on Genetic Algorithms(GAs) for speaker recognition using short speech. The GAs use the idea of heredity and evolution in biological genetics and may be viewed as an evolutionary process where in a population of solutions evolves over a sequence of generations to locate the robust global solutions. Through GAs, the system uses speaker's short speech to produce his/her optimal speech templates quickly. The experimental results show that this approach not only reduces the speech data and be favorable to build the speaker's templates, but also produces more robust templates than the traditional methods.

关 键 词：话者确认 话者模板 遗传算法 评价函数 语音识别 

分 类 号：TN912.34[电子电信—通信与信息系统]