LiteVSR: Efficient Visual Speech Recognition by Learning from Speech Representations of Unlabeled Data

no code implementations • 15 Dec 2023 • Hendrik Laux, Emil Mededovic, Ahmed Hallawa, Lukas Martin, Arne Peine, Anke Schmeink

This paper proposes a novel, resource-efficient approach to Visual Speech Recognition (VSR) leveraging speech representations produced by any trained Automatic Speech Recognition (ASR) model.

Automatic Speech Recognition Automatic Speech Recognition (ASR) +2