Behavioural speaker identification: A forensic application

Steven J. Aiken, Donald G. Jamieson, Vijay Parsa

Abstract


Behavioural speaker identification refers to the process of identifying an individual as the speaker of a given utterance, based solely on auditory perception. The present study applied behavioural speaker identification to evaluate the hypothesis that a particular individual had produced a set of utterances recorded in a police telephone tap. The individual admitted producing some of these utterances but denied producing others. Forty voice samples were extracted from 16 telephone calls recorded by the police, and one call with the individual, recorded by the experimenters. Two listeners rated the similarity of these samples in a paired-comparison task. Utterance pairs were grouped into pairs of potential speakers. An ANOVA was performed on the paired-comparison ratings with potential speaker pair and rater as factors. The effect of potential speaker pair was significant, suggesting that at least some of the utterances had been produced by different speakers. Utterance pairs in which both utterances were denied by the individual were rated as most similar, while utterance pairs in which only one utterance was denied by the individual were rated as least similar. A cluster analysis revealed two distinct clusters. All utterances denied by the individual fell into one cluster, while the other cluster was comprised of all utterances which the individual admitted producing, along with the recording of the individual's voice obtained by the experimenters. Overall, the evidence suggested that the individual was accurate in identifying which utterances he had produced.

Keywords


Audition; Feature extraction; Statistical methods; Analysis of variance (ANOVA); Behavioral speaker identification

Full Text:

PDF

Refbacks

  • There are currently no refbacks.