ODYSSEY 2004 - The Speaker and Language Recognition Workshop

May 31 - June 3, 2004
Toledo, Spain

An Interpretation Framework for the Evaluation of Evidence in Forensic Automatic Speaker Recognition with Limited Suspect Data

Filippo Botti (1), Anil Alexander (2), Andrzej Drygajlo (2)

(1) Institut de Police Scientifique (IPS) - University of Lausanne (UNIL)
(2) Signal Processing Institute - Swiss Federal Institute of Technology (EPFL) Lausanne, Switzerland

This paper proposes an interpretation framework for the evaluation of evidence in forensic automatic speaker recognition where only a single recording of the suspect and a single trace (questioned recording) are provided. In such a case the withinsource variability of the suspect cannot be evaluated. An estimation of within-source and between-sources variability is performed for the case using a database of speakers recorded in the similar conditions of the case. Two measures of interpreting the evidence within the Bayesian framework are compared, one using the Likelihood Ratio (LR) and the other, giving complementary information, using Error Ratio (ER). Experimental results using the proposed methodology are presented and the two interpretation measures are discussed.

Full Paper

Bibliographic reference.  Botti, Filippo / Alexander, Anil / Drygajlo, Andrzej (2004): "An interpretation framework for the evaluation of evidence in forensic automatic speaker recognition with limited suspect data", In ODYS-2004, 63-68.