Aging effects on voice features used in forensic speaker comparison

This article assesses the impact of long-term non-contemporaneity between recordings on the strength of forensic voice evidence and provides recommendations for casework. Analyses of longitudinal recordings from a British television documentary series illustrate the effects of aging on forensically useful acoustic parameters. Recordings from eight speakers span five seven-year intervals, between ages 21 and 49. The frequency of the first three formants of nine monophthongs and two diphthongs decrease over time. Strength of evidence estimates for non-contemporaneous comparisons are calculated from these data using a likelihood ratio approach. Longer delays predictably result in weaker and fewer correct LRs. The effect of aging on the performance of an ASR system (BATVOX) is tested; performance varies between speakers, but deteriorates considerably in longer delays for all speakers. Findings from this and similar studies should be considered when carrying out formant- or ASR-based comparisons across long delays, and in selecting age-appropriate reference data.