3D Facial Thai Speech Animation Generation for Game Development

This paper presents a method for generating 3D facial animations of characters speaking Thai, with the objective of augmenting realism in game production. The main issue is the lack of tools and datasets specifically designed for Thai phonetic features, which differ from other languages in terms of sound structure and articulation. To resolve this issue, we created a plugin for Unreal Engine that uses phonetic data to generate realistic face motions, using the TSynC-3miti as the main sources for model training. With particular attention to the phonetic needs of the Thai language, facial motions are produced by combining a sliding window approach for temporal sequence prediction with an ensemble decision tree model. The plugin takes Thai text as input, converts it into phonetic symbols, and generates synchronized face motions for a MetaHuman 3D character model in Unreal Engine. Testing showed that our approach generates more realistic animations than Motion capture and Audio driven method, whether viewed from the front or at an angle. Our model delivers a more precise representation of pronunciation across different phonetic symbols. This capability makes it well-suited for broader applications in interactive media and customized content designed for Thai-speaking audiences.