Toward Shifting to a Verbal Likelihood Ratio Framework for Forensic Voice Comparison at the BKA Text, Speech, and Audio Department: Part 2—Common/Specific Source and Score-/Feature-Based

This is Part 2 of a report outlining a strategy to change the conclusion framework in forensic voice comparison from verbal posterior probabilities to verbal likelihood ratios at the Text, Speech and Audio Department of the German Bundeskriminalamt. Like in Part 1, this part of the report addresses how to combine results from different methods when these methods have different preferences, this time with respect to the distinctions between common source versus specific source and score-based versus feature-based. In particular, the common/specific source dichotomy leads to different ways of stating the two mutually exclusive hypotheses necessary for the verbal likelihood ratios, yet combinations of evidence in the Bayesian framework require the hypotheses to be identical. Three different solutions to this combination problem are offered, one based on methodological considerations, one on the case scenario (trace–trace or trace–reference), and the third one on semantic reasoning. The first of the solutions is preferred, based on congruence with the decision process in Part 1 of this report. Before addressing this combination issue, it is demonstrated how the common versus specific source and score- versus feature-based dichotomies constitute a four-way typology, and it is shown how different voice comparison methods can be located in this typology. For some of the methods, the location is fixed by design, and in others, there is more flexibility.