Forensic voice comparison (FVC) involves a comparative analysis of suspect and offender speech data for use in a court of law, the Bayesian Likelihood-ratio (LR) Framework being one approach for undertaking such comparisons. Within this framework there are a number of strategies that have been proposed. Assessing the strengths and weaknesses of each requires undertaking a large number of FVC experiments using speech extracted from a forensic speech database. Of concern is the fact that the results of such experiments can exhibit a high degree of variability, this being linked to the specific speakers allocated to each of the Testing, Development and Background data sets associated with an experiment. The purpose of this project is to investigate this variability and the extent to which it impacts on the reliability of a FVC.
Lab allocations have not been finalised