Audio-Visual Representation Learning via Knowledge Distillation from Speech Foundation Models

Open paper

Benchmark	Model	Rank	Results
audio-visual-speech-recognition-on-lrs3-ted	DistillAV	#5	Word Error Rate (WER): 1.3
lipreading-on-lrs3-ted	DistillAV	#8	Word Error Rate (WER): 26.2