Speaker Identification in Whisper

Sociophonetic methods and findings have value in application to real-life issues, including providing expert forensic evidence in legal cases. Forensic cases often involve voices which differ markedly from those typically encountered in laboratory or field studies. We assess the ability of people to identify familiar voices produced in whisper, a commonly used form of disguise. Members of a pre-existing social network were recorded speaking normally and in whisper. Speakers found it difficult to maintain whisper beyond 30 seconds. They and other members of the group listened to extracts that were (i) short and whispered, (ii) long and whispered, and (iii) short and normal (non-whispered). Foils were also included. Performance was well above chance, and improved significantly in conditions (ii) and (iii). Differences were found across listeners and voices. The study emphasises how important is it not to overgeneralise from experimental data to a witness’s ability under forensic conditions.