说话人识别综述  

A Survey of Speaker Recognition

作　　者：张葛祥 曾鑫 姚光乐[2] 杨强[1] 贾勇[3] 朱明[1] ZHANG Gexiang;ZENG Xin;YAO Guangle;YANG Qiang;JIA Yong;ZHU Ming(School of Automation,Chengdu University of Information Technology,Chengdu 610225,China;a.College of Computer and Network Security,Chengdu University of Technology,Chengdu 610059,China;College of Mechanical and Electrical Engineering,Chengdu University of Technology,Chengdu 610059,China)

机构地区：[1]成都信息工程大学自动化学院,四川成都610225 [2]成都理工大学计算机与网络安全学院,四川成都610059 [3]成都理工大学机电工程学院,四川成都610059

出　　处：《控制工程》2025年第2期251-264,共14页Control Engineering of China

基　　金：国家自然科学基金资助项目(61972324);四川省科技项目(2021YFS0313,23NSFTD0049,2021YFG0133,23ZDYF0247)。

摘　　要：说话人识别是一种利用人类语音中存在的与身份相关的特征来识别说话人身份的技术,也称为声纹识别。这项技术既是新一代人工智能的重要研究方向,也是计算机领域的研究热点,吸引了许多学者的关注与研究。首先,在结合以往和当前有关该领域的研究工作基础上,概括性地介绍了说话人识别的重要背景知识;然后,综述了以往工作,并从数据集、评估指标和相关比赛三部分对说话人识别系统性能评估进行了一个详细的梳理和归纳;最后,指出了在当前说话人识别方法中存在的难点与挑战。Speaker recognition,also called voiceprint recognition,refers to the identification of speaker identity with the characteristics related to speaker identity existing in voice.It is one of the important research directions of new generation artificial intelligence and also one of the research hotspots in the field of computer science,which has attracted much attention.On basis of the work in the literature,the important background knowledge on speaker recognition is reviewed.Then,the past work is summarized and the performance evaluation of speaker recognition system is summarized in detail from three parts:data set,evaluation index and related competition.Finally,the difficulties and challenges in the existing speaker recognition methods are pointed out.

关 键 词：说话人识别 声纹识别 说话人欺诈 深度学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TN912.34[自动化与计算机技术—控制科学与工程]