Frame Pairwise Distance Loss for Weakly-supervised Sound Event Detection

TL;DR

This paper introduces a novel Frame Pairwise Distance loss for weakly-supervised sound event detection, improving recognition accuracy with minimal labeled data and synthesized samples, validated on the DCASE 2023 dataset.

Contribution

It proposes a new FPD loss branch and sampling strategies to enhance weakly-supervised sound event detection performance.

Findings

Improved detection accuracy on DCASE 2023 dataset

Effective use of synthesized data with minimal annotations

Validation of two distinct distance metrics

Abstract

Weakly-supervised learning has emerged as a promising approach to leverage limited labeled data in various domains by bridging the gap between fully supervised methods and unsupervised techniques. Acquisition of strong annotations for detecting sound events is prohibitively expensive, making weakly supervised learning a more cost-effective and broadly applicable alternative. In order to enhance the recognition rate of the learning of detection of weakly-supervised sound events, we introduce a Frame Pairwise Distance (FPD) loss branch, complemented with a minimal amount of synthesized data. The corresponding sampling and label processing strategies are also proposed. Two distinct distance metrics are employed to evaluate the proposed approach. Finally, the method is validated on the DCASE 2023 task4 dataset. The obtained experimental results corroborated the efficacy of this approach.