Audio Barlow Twins: Self-Supervised Audio Representation Learning

TL;DR

Audio Barlow Twins introduces a self-supervised learning method for audio that adapts the Barlow Twins approach from vision, achieving state-of-the-art results on diverse audio tasks without negative samples.

Contribution

It presents a novel adaptation of Barlow Twins for audio, enabling effective self-supervised learning on large-scale audio data without negative samples.

Findings

Outperforms existing self-supervised audio methods on multiple tasks.

Achieves results comparable to state-of-the-art supervised methods.

Demonstrates effectiveness of the Barlow Twins approach in the audio domain.

Abstract

The Barlow Twins self-supervised learning objective requires neither negative samples or asymmetric learning updates, achieving results on a par with the current state-of-the-art within Computer Vision. As such, we present Audio Barlow Twins, a novel self-supervised audio representation learning approach, adapting Barlow Twins to the audio domain. We pre-train on the large-scale audio dataset AudioSet, and evaluate the quality of the learnt representations on 18 tasks from the HEAR 2021 Challenge, achieving results which outperform, or otherwise are on a par with, the current state-of-the-art for instance discrimination self-supervised learning approaches to audio representation learning. Code at https://github.com/jonahanton/SSL_audio.