Experiments with Two Forensic Automatic Speaker Comparison Systems Using Reference Populations that (Mis)Match the Test Language

Experiments were conducted, using two Forensic Automatic Speaker Comparison-systems (FASC), in which the language spoken in the Reference Population (RP) matches or mismatches the language spoken by perpetrator and suspect in a corpus of same-source and different-source pairs from NFI-FRITS. The LLRs are generally higher under mismatch between test and RP than under match (right-shift-effect). The existence and magnitude of this pattern varies depending on the test language, the automatic system, and on whether the same-source or the different-source distribution is examined. Possible explanations of the findings and practical implications for forensic casework, are discussed. It is argued that FASC can be applied in casework with language-mismatching RPs, but the right shift effect has to be taken into account.