Forensic speaker recognition with BA-LR: calibration and evaluation on a forensically realistic database - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2024

Forensic speaker recognition with BA-LR: calibration and evaluation on a forensically realistic database

Résumé

The Likelihood Ratio (LR) is fundamental in presenting foren- sic speaker recognition (FSR) results. Despite its theoretical benefits, conventional LR estimation lacks transparency, imped- ing courtroom reliability assessment. In response, the Binary- Attribute-based Likelihood Ratio (BA-LR) framework models speech extracts based on the presence or absence of a set of speaker-specific attributes. It estimates the LR as a function of attribute-based LRs. Previous works demonstrated BA-LR’s three levels of interpretability: explicit computation of attribute- based LRs, explicit contribution of these LRs to the final LR and phonetic description of the attributes, promising a fully trans- parent FSR solution. This work adds an examination of LR cal- ibration using a forensically realistic database. Logistic regres- sion is used for calibration purposes, as well as for a regularized fusion of attribute-Log LRs. Results highlight robustness and generalization ability of BA-LR, particularly in forensics.
Fichier principal
Vignette du fichier
Odyssey2024_camera_ready.pdf (467.41 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04630349 , version 1 (01-07-2024)

Identifiants

  • HAL Id : hal-04630349 , version 1

Citer

Imen Ben-Amor, Jean-François Bonastre, David van der Vloed. Forensic speaker recognition with BA-LR: calibration and evaluation on a forensically realistic database. Odyssey 2024, Jun 2024, Québec, Canada. ⟨hal-04630349⟩

Collections

UNIV-AVIGNON LIA
0 Consultations
0 Téléchargements

Partager

Gmail Mastodon Facebook X LinkedIn More