ElasticAST: An Audio Spectrogram Transformer for All Length and Resolutions

TL;DR

ElasticAST introduces a flexible transformer-based audio spectrogram model that handles variable-length inputs during training and inference, maintaining high performance across different audio lengths and resolutions.

Contribution

It proposes a novel sequence packing method enabling AST models to process variable-length audio inputs during both training and inference.

Findings

ElasticAST performs comparably to fixed-length ASTs across various lengths.

It achieves better results when trained and evaluated on native-length audio datasets.

The method maintains high accuracy across different audio resolutions.

Abstract

Transformers have rapidly overtaken CNN-based architectures as the new standard in audio classification. Transformer-based models, such as the Audio Spectrogram Transformers (AST), also inherit the fixed-size input paradigm from CNNs. However, this leads to performance degradation for ASTs in the inference when input lengths vary from the training. This paper introduces an approach that enables the use of variable-length audio inputs with AST models during both training and inference. By employing sequence packing, our method ElasticAST, accommodates any audio length during training, thereby offering flexibility across all lengths and resolutions at the inference. This flexibility allows ElasticAST to maintain evaluation capabilities at various lengths or resolutions and achieve similar performance to standard ASTs trained at specific lengths or resolutions. Moreover, experiments demonstrate ElasticAST's better performance when trained and evaluated on native-length audio datasets.