Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Energy Consumption Trends in Sound Event Detection Systems

Constance Douwes, Romain Serizel

TL;DR

A shift towards more energy-efficient approaches during training without compromising performance is highlighted, while the number of operations and the system complexity continue to grow.

Abstract

Deep learning systems have become increasingly energy- and computation-intensive, raising concerns about their environmental impact. As organizers of the Detection and Classification of Acoustic Scenes and Events (DCASE) challenge, we recognize the importance of addressing this issue. For the past three years, we have integrated energy consumption metrics into the evaluation of sound event detection (SED) systems. In this paper, we analyze the impact of this energy criterion on the challenge results and explore the evolution of system complexity and energy consumption over the years. We highlight a shift towards more energy-efficient approaches during training without compromising performance, while the number of operations and system complexity continue to grow. Through this analysis, we hope to promote more environmentally friendly practices within the SED community.

Energy Consumption Trends in Sound Event Detection Systems

TL;DR

A shift towards more energy-efficient approaches during training without compromising performance is highlighted, while the number of operations and the system complexity continue to grow.

Abstract

Deep learning systems have become increasingly energy- and computation-intensive, raising concerns about their environmental impact. As organizers of the Detection and Classification of Acoustic Scenes and Events (DCASE) challenge, we recognize the importance of addressing this issue. For the past three years, we have integrated energy consumption metrics into the evaluation of sound event detection (SED) systems. In this paper, we analyze the impact of this energy criterion on the challenge results and explore the evolution of system complexity and energy consumption over the years. We highlight a shift towards more energy-efficient approaches during training without compromising performance, while the number of operations and system complexity continue to grow. Through this analysis, we hope to promote more environmentally friendly practices within the SED community.
Paper Structure (14 sections, 2 figures, 6 tables)

This paper contains 14 sections, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Analysis setup and novelties
  3. Task setup and performance metrics
  4. Energy and compute metrics
  5. Submissions
  6. Energy and complexity evolution
  7. General comparisons (2022 - 2024)
  8. Energy normalized comparisons (2023 - 2024)
  9. GPU energy consumption (2024)
  10. Performance analysis
  11. Energy and performance comparisons (2023 - 2024)
  12. Energy-weighted performance (2024)
  13. Threshold based on energy consumption (2023 - 2024)
  14. Conclusions

Figures (2)

  • Figure 1: Relationship between PSDS and training energy consumption for the best ensemble and non-ensemble systems from 2023 and 2024. Baselines from both years are reported as references.
  • Figure 2: Relationship between energy-weighted (EW) metrics and performance (PSDS and segMPAUC) for ensemble and non-ensemble systems in 2024, compared to the baseline system.