Skip to Main content Skip to Navigation

Vérification de l'identité d'un visage parlant.
Apport de la mesure de synchronie audiovisuelle face aux tentatives délibérées d'imposture.

Abstract : Authenticating people is a means to ensure the safety of people, goods or sensitive information, which is one of the major concerns of our modern societies. The main weakness of current authentication systems is that they rely on what a person owns (a passport, a magnetic card, etc.) and/or what he/she knows (a PIN number, a password, etc.). Still, a card can be stolen and a password broken.

Biometrics is the technological field dealing with authentication and/or identification of people using their physical characteristics, including morphological and behavioral measurements. This happens to be an obvious solution to the issue previously highlighted : the identity of a person is then related to who he/she is and no longer to what he/she owns or knows.

Not only is talking face one of the less intrusive biometric modality, it is also intrinsically multimodal : it includes both speaker and face verification, and a third modality related to audiovisual speech synchrony between the voice and lip motion.

In the first part of this document, we overview the literature about the talking-face biometric modality and we underline that deliberate impostor attacks are often forgotten in the development process of talking-face authentication algorithms. We simulate four kinds of deliberate impostor replay attacks in order to uncover the main weakness of classical systems based on the fusion of speaker and face verification scores.

In the second part, we propose to study the audiovisual synchrony between voice and lip motion as a way of making a classical speaker+face robust to attacks. Several novel synchrony measures based on canonical correlation analysis and co-inertia analysis are introduced and tested on the asynchrony detection task. The promising results that we obtained with a client-dependent synchrony measure led us to introduce a novel biometric modality based on audiovisual synchrony. Though it is not as efficient as speaker and face verification, this new modality is intrinsically robust to deliberate impostor attacks. We finally point out the complementarity between the speaker+face reference system and the synchrony modality and introduce novel fusion strategies that allow to achieve a good compromise between the efficiency of the former and
the robustness of the latter.
Complete list of metadata
Contributor : Hervé Bredin Connect in order to contact the contributor
Submitted on : Wednesday, March 19, 2008 - 2:20:41 PM
Last modification on : Wednesday, November 4, 2020 - 10:08:02 PM
Long-term archiving on: : Thursday, May 20, 2010 - 10:16:30 PM


  • HAL Id : tel-00265505, version 1



Hervé Bredin. Vérification de l'identité d'un visage parlant.
Apport de la mesure de synchronie audiovisuelle face aux tentatives délibérées d'imposture.. Traitement du signal et de l'image [eess.SP]. Ecole nationale supérieure des telecommunications - ENST, 2007. Français. ⟨tel-00265505⟩



Record views


Files downloads