Publications & Research
Peer-reviewed publications, conference papers, and research collaborations in biometrics, deepfake detection, and synthetic voice security.
DeepFake Detection in Dyadic Video Calls using Point of Gaze Tracking
IEEE Access · 2026This work presents a real-time deepfake detection approach for dyadic video calls using point-of-gaze tracking as a biometric cue. The method is built on explainable gaze-based features and evaluated on a novel dataset, achieving 82% accuracy. It is among the first reported approaches to use point-of-gaze tracking for real-time deepfake detection in conversational settings.
View PreprintDeepfake Attacks on Biometric Recognition: Evaluation of Resistance to Injection Attacks
IEEE UEMCON 2024 · Presented at IBM T.J. Watson Research CenterThis paper studies how biometric systems respond to injection attacks using deepfake face and voice media. It analyzes the vulnerability of recognition pipelines under realistic attack conditions and highlights the need for stronger liveness and anti-spoofing defenses in biometric authentication systems.
View PublicationFace Liveness Detection Competition (LivDet-Face) - 2024
IEEE International Joint Conference on Biometrics (IJCB) · 2024This publication reports the results of LivDet-Face 2024, a benchmark competition for face presentation attack detection. It compares modern anti-spoofing systems using a common evaluation protocol and standardized datasets, helping advance robust face liveness detection research.
View PublicationDescriptor: Extended-Length Audio Dataset for Synthetic Voice Detection and Speaker Recognition (ELAD-SVDSR)
IEEE Data / Signal Processing Society · 2026This paper presents ELAD-SVDSR, a large-scale extended-duration audio dataset designed for synthetic voice (deepfake) detection and speaker recognition. The dataset includes ~45-minute recordings from 36 participants captured using five distinct microphones, along with 20 deepfake voice samples. By focusing on long-duration audio, the dataset captures nuanced speech characteristics such as pitch variation, intonation, and delivery patterns, enabling more robust deepfake generation and detection models. ELAD-SVDSR addresses limitations in existing datasets by providing extended recordings, microphone diversity, and integrated synthetic samples, making it a valuable resource for audio forensics, biometric security, and voice authentication research.
View Publication