Biometric Evidence in Forensic Automatic Speaker Recognition

• Unité de recherche: COST
• Numéro de projet: C12.0117
• Titre du projet anglais: Biometric Evidence in Forensic Automatic Speaker Recognition

Mots-clé (Anglais)

evidence; recognition; speaker; biometrics; forensics; quality

Programme de recherche (Anglais)

COST-Action IC1106 - Integrating Biometrics and Forensics for the Digital Age

Description succincte (Anglais)

The major scientific and technological aspect in the domain of forensic biometrics is that there is a critical need for developing biometric recognition methods in the light of current state-of-the-art biometric technology related to the interpretation of forensic evidence. The goal of this project is the definition and development of automatic systems for forensic speaker recognition (FSR) based on scientifically approved biometric methods for calculation and interpretation of forensic evidence and validated on multilingual databases recorded in real-life conditions or collected with forensic application in mind. The final objective of this project is not to promote one method against another, but is to make available to the legal and investigative bodies in Switzerland and Europe, in particular those belonging to the European Network of Forensic Science Institutes (ENFSI), forensic tools based on the scientific approach and a rigorous experimental background.

Autres indications (Anglais)

Full name of research-institution/enterprise: EPF Lausanne Head of the Biometrics and Speech Processing Group EPFL-STI-IEL-LIDIAP, ELD 230

Partenaires et organisations internationales (Anglais)

AT; BE; HR; CY; CZ; DK; FI; FR; DE; EL; IE; IL; IT; LV; MT; NL; NO; PL; PT; SK; SI; ES; SE; CH; MK; TR; UK

Résumé des résultats (Abstract) (Anglais)

The major scientific and technological aspect in the domain of forensic biometrics is that there is a critical need for developing biometric recognition methods in the light of current state-of-the-art biometric technology related to the interpretation of forensic evidence. Forensic speaker recognition is the process of determining if a specific individual (suspected speaker) is the source of a questioned speech recording (trace). This project aims at developing forensic automatic speaker recognition (FASR) methods and methodology that provide a coherent way of quantifying and presenting recorded speech as biometric evidence, as well as the assessment of its strength (likelihood ratio) in the Bayesian interpretation framework compatible with interpretations in other forensic disciplines. Within the second year period of the project, the scientific and technological objectives were addressed along one central innovative axis of research and development: Comparative analysis and evaluation of developed methods and techniques using simulated and real case databases. More specifically, the project addressed the following main issues: 1) analysis of the relevance and the formulation of hypotheses in the framework of Bayesian interpretation using scoring (univariate) and direct (multivariate) modelling of the hypotheses, 2) Assessment of the performance of multi-variate and uni-variate systems using a variety of acoustic feature vectors in combination with deterministic and statistical speaker models and their comparison with other systems under development, 3) Evaluation and validation of developed methods and techniques using simulated and real case databases. The developed forensic automatic speaker recognition methodology has proven an effective tool in the fight against crime, yet there is a constant need for more research due to the difficulties involved because of the within-speaker (within-source) variability, between-speakers (betweensources) variability, and differences in recording sessions conditions. The developed forensic automatic speaker recognition methodology has proven an effective tool in the fight against crime, yet there is a constant need for more research due to the difficulties involved because of the within-speaker (within-source) variability, between-speakers (between-sources) variability, and differences in recording sessions conditions.

Références bases de données (Anglais)

Swiss Database: COST-DB of the State Secretariat for Education and Research Hallwylstrasse 4 CH-3003 Berne, Switzerland Tel. [phone] Swiss Project-Number: C12.0117