Improved Zero-Shot Audio Tagging & Classification with Patchout Spectrogram Transformers

TL;DR

This paper demonstrates that self-attention-based spectrogram transformers significantly improve zero-shot audio tagging and classification across multiple datasets compared to convolutional models.

Contribution

It introduces the patchout spectrogram transformer for zero-shot audio tasks and compares its performance with traditional convolutional architectures, showing superior results.

Findings

Self-attention models outperform convolutional models in zero-shot audio tasks.

Performance drops when semantic distance between training and test classes increases.

Patchout spectrogram transformer achieves state-of-the-art results on multiple benchmarks.

Abstract

Standard machine learning models for tagging and classifying acoustic signals cannot handle classes that were not seen during training. Zero-Shot (ZS) learning overcomes this restriction by predicting classes based on adaptable class descriptions. This study sets out to investigate the effectiveness of self-attention-based audio embedding architectures for ZS learning. To this end, we compare the very recent patchout spectrogram transformer with two classic convolutional architectures. We evaluate these three architectures on three tasks and on three different benchmark datasets: general-purpose tagging on AudioSet, environmental sound classification on ESC-50, and instrument tagging on OpenMIC. Our results show that the self-attention-based embedding methods outperform both compared convolutional architectures in all of these settings. By designing training and test data accordingly, we observe that prediction performance suffers significantly when the `semantic distance' between training and new test classes is large, an effect that will deserve more detailed investigations.