Inter-speaker variability in forensic voice comparison: A preliminary evaluation

摘要: In forensic voice comparison, it is strongly recommended to follow Bayesian paradigm. this paradigm, the strength of evidence summarized by a likelihood ratio (LR). The LR magnitude quantifies evidence: far from unity for meaningful (a which supports one hypothesis); close when next useless. Despite nice theoretical aspect, does not embed reliability its estimation process itself. And, in various cases, lack inside able destroy resulting LR. It particularly true comparison considered, as Speaker Recognition (SR) systems are outputting score all situations regardless case specific conditions. Furthermore, SR use different normalization steps see their scores and these clearly potential source bias. Consequently, complete view should be taken into account comparison. This article focuses on part question, "speaker factor", characteristics behaviors two speakers involved trial.