Large vocabulary audio-visual speech recognition using the janus speech recognition toolkit

Abstract

This paper describes audio-visual speech recognition experiments on a multi-speaker, large vocabulary corpus using the Janus speech recognition toolkit. We describe a complete audio-visual speech recognition system and present experiments on this corpus. By using visual cues as additional input to the speech recognizer, we observed good improvements, both on clean and noisy speech in our experiments. © Springer-Verlag 2004.