Shared Multi-modal Embedding Space for Face-Voice Association
This addresses the problem of cross-modal biometric association for multilingual applications, but appears incremental as it builds on existing embedding and loss techniques.
The paper tackled the FAME 2026 challenge tasks of training face-voice associations in a multilingual setting, including testing on unseen languages, by using separate uni-modal pipelines with general and demographic feature extraction, projecting features into a shared embedding space with Adaptive Angular Margin loss. The approach achieved first place with an average Equal-Error Rate of 23.99%.
The FAME 2026 challenge comprises two demanding tasks: training face-voice associations combined with a multilingual setting that includes testing on languages on which the model was not trained. Our approach consists of separate uni-modal processing pipelines with general face and voice feature extraction, complemented by additional age-gender feature extraction to support prediction. The resulting single-modal features are projected into a shared embedding space and trained with an Adaptive Angular Margin (AAM) loss. Our approach achieved first place in the FAME 2026 challenge, with an average Equal-Error Rate (EER) of 23.99%.