Recent AAI models rely on costly and limited EMA data. We propose ArtBoost, which extracts pseudo articulatory trajectories from visible facial anchors in large-scale speech-mesh datasets for pre-training, followed by fine-tuning on real EMA data. Experiments show consistent PCC and RMSE improvements across different AAI models, demonstrating that speech-mesh data can provide scalable articulatory supervision for AAI.

Hyungkyu Kim received the B.S. degree from Konkuk University, Seoul, South Korea, in 2023, and the M.S. degree from Chung-Ang University, Seoul, South Korea, in 2025. He is currently pursuing the Ph.D. degree at the Graduate School of Advanced Imaging Science, Multimedia & Film (GSAIM), Chung-Ang University, Seoul, South Korea. His research interests include speech signal processing, multimodal learning, and 3D generative models.