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

Authors: Xueping Zhang, Han Yin, Yang Xiao, Lin Zhang, Ting Dang, Rohan Kumar Das, Ming Li

Published: 2026-06-09 12:42:14+00:00

Comment: Accepted to 2026 ICME workshop

AI Summary

The ESDD2 Challenge evaluated systems for environment-aware, component-level audio spoofing detection, where speech and environmental sounds can be independently or jointly manipulated. This paper analyzes the results, highlighting effective design choices from top submissions such as modular task decomposition, cross-domain self-supervised encoders, and selective ensembling. While the best system substantially outperformed the baseline, challenges remain in detecting spoofed environmental components and generalizing to unseen generators.

Abstract

The Environment-Aware Speech and Sound Deepfake Detection Challenge (ESDD2), held in conjunction with ICME 2026, evaluated systems for five component-level audio spoofing detection, where speech and environmental sounds may be manipulated independently or jointly. After the challenge concludes, we analyze the final leaderboard and summarize effective design choices from the top-performing submissions. The challenge attracted 94 registrations from 16 countries; after verification of submission requirements and metadata, 13 teams were retained for the final analysis. On the test set, the best system achieved a Macro-F1 score of 0.8775, substantially outperforming the separation-enhanced joint learning baseline (0.6327). Top systems consistently benefited from modular task decomposition, cross-domain self-supervised encoders, targeted data augmentation, and selective ensembling rather than simple model scaling. At the same time, auxiliary EER analyses reveal persistent difficulty in detecting the spoofed environmental component and in generalizing to unseen generators in the test set. This paper reports challenge results and provides insights for future environment-aware deepfake detection research. The CompSpoofV2 dataset and baseline code remain publicly available for reproducibility.

Key findings

The best system achieved a Macro-F1 score of 0.8775, significantly surpassing the separation-enhanced joint learning baseline (0.6327). Top systems consistently benefited from modular task decomposition, cross-domain self-supervised encoders, targeted data augmentation, and selective ensembling. However, the analysis revealed persistent difficulty in detecting spoofed environmental components and in generalizing to unseen generators in the test set.

Approach

The challenge involved classifying audio clips into five classes based on whether speech and environmental sound components are bona fide or spoofed. Top-performing systems achieved this by employing modular task decomposition, leveraging diverse cross-domain self-supervised encoders, applying targeted data augmentation like RawBoost, and utilizing selective ensemble and fusion strategies.

Datasets

CompSpoofV2

Model(s)

Separation-enhanced joint learning framework (baseline), XLS-R, EAT, SSLAM, Dasheng, DF-Arena, XLSR-Mamba, SLS, TCM-ADD, AASIST

Author countries

China, South Korea, Australia, USA, Singapore

← Previous