3D Talking Heads: Advancing Realism and Generalization

This thesis addresses the problem of generating realistic and generalizable 3D talking heads from speech. It covers contributions spanning speaker-agnostic 3D facial animation, emotional and expressive talking head generation, animation of unregistered meshes with arbitrary topologies, and improved evaluation metrics for lip-syncing and 3D face reconstruction, advancing both the realism and the generalization capabilities of speech-driven 3D facial animation methods.