Looking into the real world: Likelihood ratio variability under forensically realistic conditions

This study sets out to investigate how the speech of a single speaker can vary depending on their interlocutor and apparent emotional status and, consequently, how this affects likelihood ratio (LR)-based forensic voice comparison, using authentic data from past forensic casework. The results revealed that these factors do have a significant impact on the LR calculations, and voice comparisons between the testing data with mismatched conditions produce much less reliable results than those recorded under matching conditions.