Learning Spatial Features from Audio-Visual Correspondence in Egocentric Videos

arXiv:2307.04760v3 Announce Type: replace
Abstract: We propose a self-supervised method for learning representations based on spatial audio-visual correspondences in egocentric videos. Our method uses a masked auto-encoding framework to synthesize masked binaural (multi-channel) audio through the synergy of audio and vision, thereby learning useful spatial relationships between the two modalities. We use our pretrained features to tackle two downstream video tasks requiring spatial understanding in social scenarios: active speaker detection and spatial audio denoising. Through extensive experiments, we show that …

arxiv audio cs.cv cs.sd eess.as features spatial type videos visual