基于似然率方法的语音证据评价

Strength Evaluation of Forensic Speaker Recognition Evidence based on Likelihood Ratio Approach

法庭话者识别技术引入我国已有二十多年,目前该技术已经广泛应用于各类刑事和民事案件中。纵观世界各国的法庭证据评价体系,除了DNA证据以外,其他证据检验鉴定结论的表述通常有三种:同一认定、否定排除和倾向性意见。简单肯定和否定的结论存在很大的弊端,即过高估计了证据的力度,因而有时可能会造成错案。鉴于此,国外一些学者提出在鉴定结论的表述和价值评判上引入基于贝叶斯理论的"似然率"方法。本文结合目前的法庭话者鉴别方法,重点介绍基于贝叶斯理论下的"似然率"理论体系,并阐明该方法的优越性、局限性以及一些亟待解决的问题。

Forensic speaker recognition was introduced into China more than 20 years ago, and has been widely used in all kinds of criminal and civil cases. As far as the evaluation of forensic evidence is concerned, conclusions are usually expressed as ＂Identification/ Elimination/ No Conclusion＂. The expression of absolute ＂YES/NO＂ decision like this has a great disadvantage： it unduly estimates the strength of evidence. This will probably result in some wrong convictions. In view of this, some forensic scientists advocate the use of the Likelihood Ratio of Bayes＇ Theorem to help to express the conclusions and evaluate the strength of forensic evidence. This paper introduces the Likelihood Ratio-based approach into forensic speaker recognition, and discusses its advantages, limitations and some unresolved problems.