Investigating Design Choices in Joint-Embedding Predictive Architectures for General Audio Representation Learning
This work addresses audio representation learning for tasks like environmental sounds, speech, and music, but it is incremental as it adapts an existing framework to a new modality.
The paper tackled self-supervised general-purpose audio representation learning by exploring Joint-Embedding Predictive Architectures (JEPA), investigating design choices like context-target splits and showing that effective image-domain designs perform poorly on audio, highlighting modality differences.
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.