Sign Language Synthesis using Pose Priors

The challenge of simulating realistic Sign Language using avatars lies in achieving accurate human-like postures for effective communication. Unlike artistic or motion capture techniques, linguist-driven procedural generation methods are widely employed, relying on skeletal representations to synthesize a broad range of signs. However, determining appropriate joint limits for these avatars is intricate due to inter-joint and intra-joint dependencies, as well as variations in biomechanical properties. In this context, our work addresses this problem by introducing a pose corrector, enhancing an established Sign Language synthesis technique. Focused on rectifying extreme joint rotations, our approach incorporates a pre-trained poser based on existing work, integrated with a 21-joint character model. The correction process involves applying linguist-defined constraints using AZee language and subsequent pose corrections, showcasing promising advancements in obtaining more natural sign gestures.