Improving Sound Event Detection Metrics: Insights from DCASE 2020

TL;DR

This paper evaluates different sound event detection metrics, highlighting the advantages of the PSDS intersection-based criterion over traditional event-based and segment-based methods in providing fairer, more robust system comparisons.

Contribution

It introduces and advocates for the PSDS intersection-based criterion, demonstrating its advantages in unbiased, robust evaluation of sound event detection systems.

Findings

Conventional event-based criteria are affected by event length.

Segment-based criteria may lack precision and be application dependent.

PSDS's intersection-based criterion offers unbiased, robust evaluation.

Abstract

The ranking of sound event detection (SED) systems may be biased by assumptions inherent to evaluation criteria and to the choice of an operating point. This paper compares conventional event-based and segment-based criteria against the Polyphonic Sound Detection Score (PSDS)'s intersection-based criterion, over a selection of systems from DCASE 2020 Challenge Task 4. It shows that, by relying on collars , the conventional event-based criterion introduces different strictness levels depending on the length of the sound events, and that the segment-based criterion may lack precision and be application dependent. Alternatively, PSDS's intersection-based criterion overcomes the dependency of the evaluation on sound event duration and provides robustness to labelling subjectivity, by allowing valid detections of interrupted events. Furthermore, PSDS enhances the comparison of SED systems by measuring sound event modelling performance independently from the systems' operating points.