Cross-Language Forensic Voice Comparison of Hong Kong Trilingual Speakers using Filled Pauses and an Automatic Speaker Recognition System

Forensic voice comparison (FVC) involving bilingual speakers presents a substantial challenge to forensic practitioners. Previous FVC research suggests that discriminatory power is, unsurprisingly, weaker in language mismatch conditions than in the language match conditions. The present paper extends on previous work examining bilingual speakers to consider trilingual speakers. Specifically, we conduct FVC tests of Cantonese-English-Mandarin trilinguals using acoustic-phonetic data extracted from filled pauses and using an automatic speaker recognition (ASR) system. A different language effect was found in the two systems, such that the Mandarin-English condition produced the best performance in the acoustic-phonetic based system, while the Cantonese-English condition produced the worst performance in the ASR system. Individual-speaker analysis suggests that trilingual speakers' proficiency in L2 and L3 seems to affect how well they perform in cross-language FVC.