SoundingActions: Learning How Actions Sound from Narrated Egocentric Videos

arXiv:2404.05206v1 Announce Type: new
Abstract: We propose a novel self-supervised embedding to learn how actions sound from narrated in-the-wild egocentric videos. Whereas existing methods rely on curated data with known audio-visual correspondence, our multimodal contrastive-consensus coding (MC3) embedding reinforces the associations between audio, language, and vision when all modality pairs agree, while diminishing those associations when any one pair does not. We show our approach can successfully discover how the long tail of human actions sound from egocentric video, outperforming …

abstract arxiv audio coding consensus cs.cv cs.mm cs.sd data eess.as embedding language learn multimodal novel sound type videos vision visual