EmoZone-Talker: Regional Semantic Control of Audio-Driven 3DGS Talking Heads via Facial Action Units
Researchers have developed EmoZone-Talker, a new framework for generating realistic 3D talking heads from audio. This system addresses the challenge of controlling facial expressions by explicitly disentangling spatial and temporal aspects of facial motion. It uses a novel approach called Synergy Zones with Prioritized Attention Bias (SZ-PAB) to manage contributions from different modalities and a Channel-Independent Temporal AU Encoder (CIT-AE) to model consistent facial action unit dynamics, leading to improved expression accuracy and temporal coherence. AI
IMPACT Introduces a novel method for more controllable and realistic facial expression synthesis in 3D talking head models.