Database selection for forensic voice comparison
Geoffrey Stewart Morrison, Felipe Agustin Ochoa, Tharmarajah Thiruvaran · 2012
Defining the relevant population to sample is an important issue in data-based implementation of the likelihood-ratio framework for forensic voice comparison. A forensic likelihood ratio is the answer to a specific question which depends on the prosecution and defence hypotheses and the circumstances of the case. If an inappropriate background sample is selected the likelihood ratio produced by the forensic-voice-comparison system will not answer the question asked, and if an inappropriate test sample is selected the results of validity and reliability tests will not be informative as to the performance of the system under conditions reflecting those of the case at trial. We present a logical argument that because an investigator or prosecutor only submits suspect and offender recordings for forensic analysis if they sound sufficiently similar to each other, the appropriate defence hypothesis for the forensic scientist to adopt will usually be that: the suspect is not the speaker on the offender recording but is a member of a population of speakers who sound sufficiently similar to the offender recording that an investigator or prosecutor would submit recordings of these speakers for forensic comparison with the offender recording. We argue that categories such as speaker gender and dialect spoken are not relevant per se. We propose a procedure for selecting background, development, and test databases using a panel of human listeners, who select voice recordings on the basis of their perceived similarity to the actual offender recordings. We briefly demonstrate