Sound Event Detection with Boundary-Aware Optimization and Inference

TL;DR

This paper introduces a boundary-aware approach for sound event detection that explicitly models event onsets and offsets, leading to more accurate and scalable detection without extensive post-processing.

Contribution

It proposes new temporal modeling layers and loss functions that improve event detection accuracy and efficiency in sound event detection tasks.

Findings

Outperforms traditional frame-wise models on AudioSet

Eliminates need for post-processing hyperparameter tuning

Achieves new state-of-the-art performance across AudioSet classes

Abstract

Temporal detection problems appear in many fields including time-series estimation, activity recognition and sound event detection (SED). In this work, we propose a new approach to temporal event modeling by explicitly modeling event onsets and offsets, and by introducing boundary-aware optimization and inference strategies that substantially enhance temporal event detection. The presented methodology incorporates new temporal modeling layers - Recurrent Event Detection (RED) and Event Proposal Network (EPN) - which, together with tailored loss functions, enable more effective and precise temporal event detection. We evaluate the proposed method in the SED domain using a subset of the temporally-strongly annotated portion of AudioSet. Experimental results show that our approach not only outperforms traditional frame-wise SED models with state-of-the-art post-processing, but also removes the need for post-processing hyperparameter tuning, and scales to achieve new state-of-the-art performance across all AudioSet Strong classes.