Natural Language Supervision for General-Purpose Audio Representations

TL;DR

This paper introduces a contrastive pretraining approach for audio-language models using 4.6 million audio-text pairs, resulting in improved zero-shot and downstream task performance across diverse audio applications.

Contribution

It proposes a novel contrastive language-audio pretraining framework with specialized encoders trained on multiple tasks, advancing general-purpose audio representation learning.

Findings

Achieved state-of-the-art results on several audio tasks.

Demonstrated strong zero-shot generalization across 26 tasks.

Improved downstream performance by leveraging diverse training data.

Abstract

Audio-Language models jointly learn multimodal text and audio representations that enable Zero-Shot inference. Models rely on the encoders to create powerful representations of the input and generalize to multiple tasks ranging from sounds, music, and speech. Although models have achieved remarkable performance, there is still a performance gap with task-specific models. In this paper, we propose a Contrastive Language-Audio Pretraining model that is pretrained with a diverse collection of 4.6M audio-text pairs employing two innovative encoders for Zero-Shot inference. To learn audio representations, we trained an audio encoder on 22 audio tasks, instead of the standard training of sound event classification. To learn language representations, we trained an autoregressive decoder-only model instead of the standard encoder-only models. Then, the audio and language representations are brought into a joint multimodal space using Contrastive Learning. We used our encoders to improve the downstream performance by a margin. We extensively evaluated the generalization of our representations on 26 downstream tasks, the largest in the literature. Our model achieves state of the art results in several tasks leading the way towards general-purpose audio representations.