APSIPA Transactions on Signal and Information Processing > Vol 13 > Issue 1

Sound Event Detection: A Journey Through DCASE Challenge Series

Tanmay Khandelwal, Fortemedia Singapore, Singapore and New York University, USA, f20170106p@alumni.bits-pilani.ac.in , Rohan Kumar Das, Fortemedia Singapore, Singapore, Eng Siong Chng, Nanyang Technological University, Singapore
 
Suggested Citation
Tanmay Khandelwal, Rohan Kumar Das and Eng Siong Chng (2024), "Sound Event Detection: A Journey Through DCASE Challenge Series", APSIPA Transactions on Signal and Information Processing: Vol. 13: No. 1, e3. http://dx.doi.org/10.1561/116.00000051

Publication Date: 13 Feb 2024
© 2024 T. Khandelwal, R. K. Das and E. S. Chng
 
Subjects
 
Keywords
Sound event detectionDCASE challengeFeature extractionPost-processingMachine learning
 

Share

Open Access

This is published under the terms of CC BY-NC.

Downloaded: 659 times

In this article:
Introduction 
SED Problem Formulation and Applications 
DCASE Challenge on SED 
Discussion and Summary 
Future Horizons 
Conclusion 
References 

Abstract

The sense of hearing is fundamental to human beings, as it allows them to perceive their surroundings. However, this simple task of recognizing different sounds in complex environments poses a challenge for machines. Sound event detection (SED) is a field that aims to automate the human auditory system’s detection and recognition of sound events with their onset and offset points. Training an SED system typically requires a large labeled set, but is associated with high annotation costs and is dependent on the subjective judgments of annotators. Therefore, significant efforts have been made in this area, including the major DCASE challenge series, which brings researchers together annually to address this issue. The DCASE challenge was started in the year 2013, and it has evolved over the years to witness some significant breakthroughs in the field of SED. In this study, we delve into the methods proposed by various authors in the DCASE challenge series, providing a thorough discussion of feature extraction, machine learning techniques, and post-processing methods. We also study the results from top teams in each edition of the DCASE challenge to bring out the highlights of the best-performing SED systems and explore potential future research directions.

DOI:10.1561/116.00000051