M2S-AVSR improves robust audio-visual speech recognition

Arxiv CS.SD (Sound/Audio ML)

M2S-AVSR introduces modality-aware multi-view self-supervised representation for robust audio-visual speech recognition, addressing challenges like viewpoint variation, audio distortion, and visual occlusion. The method leverages visual cues to enhance robustness in real-world scenarios.

self-supervised-learning

Arxiv CS.SD (Sound/Audio ML)·Fei SuCancan LiMing LiJuan Liu

6 days ago