摘要
司法实践中,检材语音与样本语音的信道往往存在不匹配情况。为了量化评估信道失配对法庭说话人识别效果的影响,本文使用60人的电话通话、呼叫中心和微信三种信道语音,对基于x-vector PLDA模型的法庭说话人自动识别系统进行了说话人识别测试,并基于似然比框架,对三种信道语言在信道匹配和失配情况下的识别性能进行了分析比较,进而量化评价信道失配对法庭说话人识别系统的影响。
In forensic practice there is often a channel mismatch between the criminal's speech samples and the suspect's speech samples.In order to quantitatively evaluate the effect of channel mismatch on forensic speaker recognition,this study conducted speaker recognition tests on a forensic automatic recognition system based on the x-vector PLDA model using 60 people's speech data from three channels of telephone calls,call center and WeChat.The performance of the system was analyzed and compared under conditions of channel match and channel mismatch within the likelihood ratio framework,and the effect of channel mismatch on forensic speaker recognition system was quantitatively evaluated.
作者
张翠玲
宋羽蕾
ZHANG Cuiling;SONG Yulei(School of Criminal Investigation,Southwest University of Political Science and Law,Chongqing 401120,China;Chongqing Institutes of Higher Education Forensic Science Key Laboratory,Chongqing,401120,China)
出处
《中国刑警学院学报》
2023年第3期121-128,共8页
Journal of Criminal Investigation Police University of China
基金
2022年度重庆市自然科学基金创新发展联合基金项目(编号:2022NSCQ-LEX0094)
2021年度西南政法大学学生科研创新项目(编号:2021XZXS-131)。
关键词
法庭说话人识别
信道失配
似然比
量化评价
forensic speaker recognition
channel mismatch
likelihood ratio
quantitative evaluation