BEAT2AASIST model with layer fusion for ESDD 2026 Challenge

TL;DR

This paper introduces BEAT2AASIST, a novel model for environmental sound deepfake detection that uses layer fusion and data augmentation to improve robustness and performance in the ESDD 2026 Challenge.

Contribution

It extends BEATs-AASIST with layer fusion techniques and vocoder-based data augmentation, enhancing feature representation and robustness for sound deepfake detection.

Findings

Achieves competitive performance on ESDD 2026 test sets.

Layer fusion improves feature representation for detection.

Data augmentation enhances robustness against unseen spoofing methods.

Abstract

Recent advances in audio generation have increased the risk of realistic environmental sound manipulation, motivating the ESDD 2026 Challenge as the first large-scale benchmark for Environmental Sound Deepfake Detection (ESDD). We propose BEAT2AASIST which extends BEATs-AASIST by splitting BEATs-derived representations along frequency or channel dimension and processing them with dual AASIST branches. To enrich feature representations, we incorporate top-k transformer layer fusion using concatenation, CNN-gated, and SE-gated strategies. In addition, vocoder-based data augmentation is applied to improve robustness against unseen spoofing methods. Experimental results on the official test sets demonstrate that the proposed approach achieves competitive performance across the challenge tracks.