What Does Your Face Sound Like? 3D Face Shape towards Voice

Face-based speech synthesis provides a practical solution to generate voices from human faces. However, directly using 2D face images leads the problems of uninterpretability and entanglement. In this paper, address issues, we introduce 3D shape which (1) has an anatomical relationship between voice characteristics, partaking in "bone conduction" timbre production, (2) is naturally independent irrelevant factors by excluding blending process. We devise three-stage framework shapes. Fully considering production acquired terms, our incorporates three additional relevant attributes including texture, facial features, demographics. Experiments subjective tests demonstrate method can utterances matching faces well, with good audio quality diversity. also explore visualize how changes face. Case studies show that upgrades face-voice inference personalized custom-made creating, revealing promising prospect virtual dubbing applications.