Investigating Design Choices in Joint-Embedding Predictive Architectures for General Audio Representation Learning

TL;DR

This paper explores how different design choices in joint-embedding architectures affect self-supervised audio representation learning, revealing significant modality-specific differences from image-based methods through extensive experiments.

Contribution

It systematically investigates the impact of various design choices in JEPA for audio, highlighting modality-specific considerations and providing insights for better audio representation learning.

Findings

Input data partitioning significantly affects model quality.

Design choices effective in images may perform poorly in audio.

Extensive experiments across diverse audio tasks validate the findings.

Abstract

This paper addresses the problem of self-supervised general-purpose audio representation learning. We explore the use of Joint-Embedding Predictive Architectures (JEPA) for this task, which consists of splitting an input mel-spectrogram into two parts (context and target), computing neural representations for each, and training the neural network to predict the target representations from the context representations. We investigate several design choices within this framework and study their influence through extensive experiments by evaluating our models on various audio classification benchmarks, including environmental sounds, speech and music downstream tasks. We focus notably on which part of the input data is used as context or target and show experimentally that it significantly impacts the model's quality. In particular, we notice that some effective design choices in the image domain lead to poor performance on audio, thus highlighting major differences between these two modalities.