Learning Spatial Features from Audio-Visual Correspondence in Egocentric Videos

TL;DR

This paper introduces a self-supervised approach that leverages spatial audio-visual correspondence in egocentric videos to learn representations, improving spatial understanding tasks like speaker detection and audio denoising.

Contribution

It presents a novel masked auto-encoding framework for synthesizing binaural audio from visual cues, enhancing spatial feature learning in egocentric videos.

Findings

Outperforms state-of-the-art baselines on active speaker detection.

Improves spatial audio denoising accuracy.

Demonstrates generalization across two challenging datasets.

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 our features are generic enough to improve over multiple state-of-the-art baselines on both tasks on two challenging egocentric video datasets that offer binaural audio, EgoCom and EasyCom. Project: http://vision.cs.utexas.edu/projects/ego_av_corr.