Presented STF-ACSR, our semi training-free cued speech recognition work, at ICASSP 2026. Paper · Code