Abstract:Voice comparison is a useful technique for forensic speaker identification.However, there is only within-language speaker identification so far.This research conducts a phonetic experiment to explore the validity of cross-linguistic voice comparison based on vowel formants.The results show that likelihood values derived from this experiment are able to identify over 50% of the tested subjects.The log-likelihood ratio costs reveal that/i/_f ₂,/ə/_f ₁and/u/_f ₁are the only three acoustic features that can meet the validity requirements for likelihood approach to voice comparison.As to the language used by the reference population, both Chinese and English are valid for/i/_f ₂and/ə/_f ₁, while only English is good for/u/_f ₁.Finally, some proposals are made on how to strengthen cross-linguistic forensic voice comparison by exploring further into the choice of phonetic features, the selection of reference population and the application of forensic voice comparison.