Calculating likelihood ratios for forensic speaker comparisons using phonetic and linguistic parameters

The research presented in this thesis examines the calculation of numerical likelihood ratios using phonetic and linguistic parameters derived from a corpus of recordings of speakers of Southern Standard British English. The research serves as an investigation into the development of the numerical likelihood ratio as a medium for framing forensic speaker comparison conclusions. The thesis begins by investigating which parameters are claimed to be the most useful speaker discriminants according to expert opinion, and in turn examines four of these ‘selected/valued’ parameters individually in relation to intra- and inter-speaker variation, their capacities as speaker discriminants, and the potential strength of evidence they yield. The four parameters analyzed are articulation rate, fundamental frequency, long-term formant distributions, and the incidence of clicks (velaric ingressive plosives). The final portion of the thesis considers the combination of the four parameters under a numerical likelihood ratio framework in order to provide an overall likelihood ratio.

The contributions of this research are threefold. Firstly, the thesis presents for the first time a comprehensive survey of current forensic speaker comparison practices around the world. Secondly, it expands the phonetic literature by providing acoustic and auditory analysis, as well as population statistics, for four phonetic and linguistic parameters that survey participants have identified as effective speaker discriminants. And thirdly, it contributes to the forensic speech science and likelihood ratios for forensics literature by considering what steps can be taken to conceptually align the area of forensic speaker comparison with more developed areas of forensic science (e.g. DNA) by creating a human-based (auditory and acoustic-phonetic) forensic speaker comparison system.