Problem
Research questions and friction points this paper is trying to address.
audio deepfake detection
bias mitigation
generalization
artifact focus
dataset bias
AFSS: Artifact-Focused Self-Synthesis for Mitigating Bias in Audio Deepfake Detection
🤖 AI Summary
Existing audio deepfake detection methods exhibit limited generalization due to their reliance on specific spoofing datasets. To address this, this work proposes a self-synthesis strategy that requires no pre-collected fake samples. By leveraging self-transformation and self-reconstruction mechanisms, the approach generates pseudo-spoofed audio while preserving speaker identity and semantic content, thereby guiding the model to focus on generation artifacts rather than irrelevant distractors. A same-speaker constraint combined with a learnable reweighted loss dynamically enhances the model’s sensitivity to synthetic traces. Evaluated across seven benchmark datasets, the method achieves an average equal error rate (EER) of 5.45%, with notably low EERs of 1.23% on WaveFake and 2.70% on In-the-Wild, significantly outperforming current state-of-the-art techniques.
📝 Abstract
The rapid advancement of generative models has enabled highly realistic audio deepfakes, yet current detectors suffer from a critical bias problem, leading to poor generalization across unseen datasets. This paper proposes Artifact-Focused Self-Synthesis (AFSS), a method designed to mitigate this bias by generating pseudo-fake samples from real audio via two mechanisms: self-conversion and self-reconstruction. The core insight of AFSS lies in enforcing same-speaker constraints, ensuring that real and pseudo-fake samples share identical speaker identity and semantic content. This forces the detector to focus exclusively on generation artifacts rather than irrelevant confounding factors. Furthermore, we introduce a learnable reweighting loss to dynamically emphasize synthetic samples during training. Extensive experiments across 7 datasets demonstrate that AFSS achieves state-of-the-art performance with an average EER of 5.45\%, including a significant reduction to 1.23\% on WaveFake and 2.70\% on In-the-Wild, all while eliminating the dependency on pre-collected fake datasets. Our code is publicly available at https://github.com/NguyenLeHaiSonGit/AFSS.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Audio Deepfake Detection
Bias Mitigation
Self-Synthesis
Generation Artifacts
Same-Speaker Constraint
🔎 Similar Papers
No similar papers found.
