Applying automatic speaker recognition systems to cross-language forensic voice comparison

Forensic voice comparison (FVC) is usually conducted to assist legal systems in forming a view of whether the recordings of a nominal suspect and those of a nominal offender are from the same individual or not. In the past decade, forensic practitioners have increasingly used automatic speaker recognition (ASR) systems, a form of AI technology, in their FVC. This paper illustrates how AI technologies impact the research and practice of FVC by reviewing the methods and approaches in FVC in the present and the past. A study of cross-language FVC is presented to showcase how ASR systems are used to tackle difficult tasks in FVC: 21 female and 22 male Cantonese–English–Mandarin trilingual speakers are tested in both language-matched and language-mismatch conditions using an x-vector ASR system. Results show that the performance of the ASR system is outstanding even when the nominal suspect and nominal offender spoke different languages (i.e., language-mismatch conditions). The paper also discusses the general impact of AI technology on forensic phonetics and brings new insights to forensic linguistics.