法庭说话人识别技术应用实证研究——以一起典型民族资产解冻诈骗案为例

An Empirical Study on the Application of Court Speaker Recognition Technology——Take a typical national asset thawing fraud case as an example

法庭说话人识别技术是利用案件中的语音进行说话人身份辨识,在一起民族资产解冻诈骗案件中,涉案语音检材数据量大、格式种类繁杂、说话人多,而且包含大量伪装语音。鉴定人员采用算法预处理、自动比对和专家鉴定相结合的方法及流程,对涉案录音证据进行了处理、分析和检验鉴定。研究表明,检材经过算法预处理后,聚类串并结果能够达到实用要求。自动比对分析结果的应用需要与检材、样本语音之间的相似度值以及检材语音在与样本条件相同数据库中的Top序列返回值相结合,两个数值综合分析可为人工检验提供量化数据支持。此外,还要注意检材语音与样本语音言语风格不匹配对自动比对结果的影响。

Court speaker recognition technology is to use the voice in the case for speaker identification,in a national assets thawing fraud case,the amount of voice examination materials involved in the case is large,the format is complex,the number of speakers is large,and it contains a large number of disguised voices.The evaluators used methods and processes combining algorithm preprocessing,automatic comparison and expert appraisal to process,analyze and examine and identify the recorded evidence involved in the case.The results show that after the pretreatment of the sample,the clustering string merging results can meet the practical requirements.The application of automatic comparison analysis results needs to be combined with the similarity value between the test material and the sample voice and the Top sequence return value of the sample voice in the database with the same sample conditions,and the two numerical comprehensive analysis can provide quantitative data support for manual testing.In addition,pay attention to the influence of the mismatch between the speech style of the sample and the speech style of the sample on the automatic comparison results.