The First Environmental Sound Deepfake Detection Challenge: Benchmarking Robustness, Evaluation, and Insights

Han Yin; Yang Xiao; Rohan Kumar Das; Jisheng Bai; Ting Dang

cs.SDMar 5, 2026

The First Environmental Sound Deepfake Detection Challenge: Benchmarking Robustness, Evaluation, and Insights

Authors: Han Yin, Yang Xiao, Rohan Kumar Das, Jisheng Bai, Ting Dang

Organizations: School of Electrical Engineering, KAIST, Daejeon, Republic of Korea · University of Melbourne, Australia · Fortemedia Singapore, Singapore · Xi’an University of Posts & Telecommunications, Xi’an, China · Xi’an Lianfeng Acoustic Technologies Co., Ltd., China

Abstract

Recent progress in audio generation has made it increasingly easy to create highly realistic environmental soundscapes, which can be misused to produce deceptive content, such as fake alarms, gunshots, and crowd sounds, raising concerns for public safety and trust. While deepfake detection for speech and singing voice has been extensively studied, environmental sound deepfake detection (ESDD) remains underexplored. To advance ESDD, the first edition of the ESDD challenge was launched, attracting 97 registered teams and receiving 1,748 valid submissions. This paper presents the task formulation, dataset construction, evaluation protocols, baseline systems, and key insights from the challenge results. Furthermore, we analyze common architectural choices and training strategies among top-performing systems. Finally, we discuss potential future research directions for ESDD, outlining key opportunities and open problems to guide subsequent studies in this field.

Figures & tables

Panel of Figure 1 — Figure 1: Overview of the two tracks fo ESDD challenge.

Usage	Real		Fake
Usage	Sources	# Clips	TTA	ATA	# Clips
Training	[ 7 , 9 , 10 , 11 ]	27,811	[ 1 , 2 , 13 ]	[ 1 ]	111,244
Validation	[ 7 , 9 , 10 , 11 ]	7,942	[ 1 , 2 , 13 ]	[ 1 ]	31,768
Test	[ 8 , 12 ]	1,000	[ 14 , 15 ]	[ 2 ]	3,000

Table 1: Statistics of the dataset in Track 1.

Usage	# Real Clips	# Fake Clips
Training	270	1,083
Validation	90	361
Test	1,994	7,980

Table 2: Statistics of the black-box data in Track 2.

Team	System	ID	Ensem	EER
AHU	EAT+AASIST	S01	5	0.30
DFKI	EAT-L+BiCrossMamba	S02	4	0.80
CUC	SSLAM+FFN	S03	1	1.20
CAU [ 20 ]	BEAT2AASIST	S04	2	1.60
CAU [ 20 ]	BEAT2AASIST	S05	2	1.62
CAU [ 20 ]	BEAT2AASIST	S06	1	1.70

Table 3: Performance in EER (%) of different detection systems in ESDD challenge Track 1. “Ensem” represents “Ensemble”.

Team	System	ID	Ensem	EER
DFKI	EAT-L-SSLAM+ BiCrossMamba	M01	5	0.25
AHU	EAT+AASIST	M02	5	0.25
CAU [ 20 ]	BEAT2AASIST	M03	3	0.35
CAU [ 20 ]	BEAT2AASIST	M04	2	0.40
CAU [ 20 ]	BEAT2AASIST	M05	3	0.41
CUC	SSLAM+FFN	M06	1	1.05

Table 4: Performance in EER (%) of different detection systems in ESDD challenge Track 2. “Ensem” represents “Ensemble”.

Generator	Method	ID	Track 1		Track 2
Generator	Method	ID	Training	Test	Training	Test
AudioLDM	TTA	G01	✔	✗	✔	✗
AudioLDM2	TTA	G02	✔	✗	✔	✗
AudioGen	TTA	G03	✔	✗	✔	✗
AudioLDM	ATA	G04	✔	✗	✔	✗
AudioLCM	TTA	G05	✗	✔	✗	✗

Table 5: Audio generators used in the ESDD challenge.

Figure 2: Deepfake detection performance in EER (%) of different systems across various audio generators in Track 1 and Track 2.

Explore similar work

Environmental Sound Deepfake Detection Using Deep-Learning Framework
Apr 21, 2026Khoi Vu, Dat Tran, Khanh Do +8Audio Deepfake Detection
FPT University, Vietnam · HCM University of Technology, Vietnam · TDT University, Vietnam +3
Overview of ESDD2: Environment-Aware Speech and Sound Deepfake Detection Challenge
Jun 9, 2026Xueping Zhang, Han Yin, Yang Xiao +4Audio Deepfake Detection
Duke Kunshan University · Korea Advanced Institute of Science and Technology · The University of Melbourne +3
EnvSSLAM-FFN: Lightweight Layer-Fused System for ESDD 2026 Challenge
Dec 23, 2025Xiaoxuan Guo, Hengyan Huang, Jiayi Zhou +5Audio Deepfake Detection
State Key Laboratory of Media Convergence and Communication, Communication University of China, Beijing, China · Machine Intelligence, Ant Group, Shanghai, China · Key Laboratory of Media Audio & Video, Ministry of Education, Communication University of China, Beijing, China

Apr 21, 2026J/K move · Enter open · S save

Environmental Sound Deepfake Detection Using Deep-Learning Framework

Khoi Vu, Dat Tran, Khanh Do, Phat Lam+7

FPT University, Vietnam · HCM University of Technology, Vietnam · TDT University, Vietnam+3

Audio Deepfake Detection

Open paper AskPDF arXiv

In this paper, we propose a deep-learning framework for Environmental Sound Deepfake Detection (ESDD) - the task of identifying whether the sound scene and sound event in an input audio recording is fake or real. To this end, we first conduct extensive experiments to explore how individual spectrograms, a wide range of network architectures, and pre-trained models affect the performance of an ESDD model. The experimental results on the benchmark datasets of EnvSDD indicate that detecting deepfake audio of sound scenes and detecting deepfake audio of sound events should be considered as individual tasks. We also show that fine-tuning a pre-trained model is more effective than training a model from scratch for ESDD. Ultimately, our best model, which fine-tunes the pre-trained BEATs model using the proposed two-phase training strategy, achieves an Accuracy of 0.98, F1 score of 0.95, and AUC score of 0.99 on the Test subset of the EnvSDD dataset. Our best model also achieves an Accuracy of 0.86, F1 score of 0.80, and AUC of 0.93 when evaluated cross-dataset on the ESD-Challenge-TestSet dataset.

Abstract

Explore similar work

Environmental Sound Deepfake Detection Using Deep-Learning Framework

Overview of ESDD2: Environment-Aware Speech and Sound Deepfake Detection Challenge

EnvSSLAM-FFN: Lightweight Layer-Fused System for ESDD 2026 Challenge