Large-Scale Weakly Labeled Semi-Supervised Sound Event Detection in Domestic Environments

TL;DR

This paper addresses large-scale sound event detection in domestic environments using weakly labeled data, focusing on improving detection accuracy by leveraging unlabeled and unbalanced datasets in a challenging real-world setting.

Contribution

It introduces methods for semi-supervised learning in sound event detection, specifically handling weak labels and unbalanced data in domestic audio recordings.

Findings

Improved detection accuracy with semi-supervised methods

Effective utilization of unlabeled data in large-scale settings

Demonstrated applicability in real-world domestic environments

Abstract

This paper presents DCASE 2018 task 4. The task evaluates systems for the large-scale detection of sound events using weakly labeled data (without time boundaries). The target of the systems is to provide not only the event class but also the event time boundaries given that multiple events can be present in an audio recording. Another challenge of the task is to explore the possibility to exploit a large amount of unbalanced and unlabeled training data together with a small weakly labeled training set to improve system performance. The data are Youtube video excerpts from domestic context which have many applications such as ambient assisted living. The domain was chosen due to the scientific challenges (wide variety of sounds, time-localized events.. .) and potential industrial applications .