Contrastive Gated Fusion for Multilingual Speaker Verification
Abstract
Understanding face-voice relationships in multilingual contexts is the focus of the FAME Challenge 2026. Unlike conventional face-voice association tasks, this challenge highlights the difficulty of transferring cross-modal face-voice models across different languages. In this work, we present a system designed to remain robust under language shifts through targeted architectural choices and training strategies. Our final submission ranked fourth in the FAME Challenge 2026, achieving an overall Equal Error Rate (EER) of 33.2 %. Our source code is publicly available at the project repository.
Fields of Study & Subjects
Face recognition and analysis
Emotion and Mood Recognition
Face Recognition and Perception
Publication details
- Journal / source
- ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
- Publisher
- IEEE
- Pages
- 21775–21777
- Publication date
- 21 April 2026