Audio-Driven Facial Animation

Audio-driven facial animation studies how speech can produce synchronized mouth movements and broader facial behavior in talking-head videos, 3D meshes, blendshapes, and Gaussian avatars. Systems use acoustic and phonetic features, often alongside text, emotion, identity, or conversational context, with Transformer, diffusion, and flow-matching models to generate motion or video. Current work aims to balance precise lip articulation with natural expression, identity preservation, personalization, and real-time or long-form generation. Evaluation remains an open challenge because frame-based metrics can miss timing differences, coarticulation, and whether motion looks natural to people.