Estimated Intra-Speaker Variability Boundaries in Forensic Speaker Recognition Casework
Yosef A. Solewicz, Gaëlle Jardine, Timo Becker, Stefan Gfroerer · 2013
Current automatic forensic speaker recognition algorithms are known to be highly dependent on matched reference data for producing statistically reliable outcomes. In day-to-day forensic casework, this data is often not available. This paper addresses this problem by introducing a data-driven modular method which does not require precise reference data. Results are more general and given as p-values and likelihood ratio ranges. Experiments performed with a real forensic database support our proposal.