Speech-Driven 3D Facial Animation Based on Diffusion Model

Speech-driven 3D facial animation is crucial in computer graphics and vision, with applications in VR, animation, telepresence, film, and gaming. Deep learning has advanced this field, but a lack of high-quality datasets has hindered model development. We introduce a large-scale 3D facial dataset with synchronized audio and detailed 3D meshes, featuring speakers of Chinese and Tibetan with professional skills. Current models often produce consistent animations, ignoring the natural randomness of non-verbal signals. To address this, we propose a probabilistic deep learning model that generates more random and natural speech-driven animations.