BA-LR: Binary-Attribute-based Likelihood Ratio estimation for forensic voice comparison
Imen Ben Amor, Jean-François Bonastre · 2022
Likelihood ratio (LR) is a widely adopted paradigm in forensic science to represent the conclusion of a practitioner report. However, with existing estimation methods, the LR does not fully facilitate the decision making by judges and juries. With an explained decomposition of the LR value together with the case information, the judge can more easily take in hands the weight of evidence in the final decision. To that end, we propose the Binary-Attribute-based LR estimation approach (BA-LR) for Forensic Voice Comparison (FVC) where the LR is obtained as the composition of partial LRs, each dedicated to an attribute. An attribute is expressed by the presence or absence of a speaker voice characteristic. The partial LRs are directly computed following a formulation of prosecution and defence hypotheses using three probabilities that describe explicitly the behavior of the considered attribute. An implementation of our approach is evaluated on VoxCeleb1&2. The results show the effectiveness of our BA-LR approach and give hope on a better handling of LR-based FVC conclusions.