AFSS: Artifact-Focused Self-Synthesis for Mitigating Bias in Audio Deepfake Detection

1. 1Generate pseudo-fake samples from real audio using self-conversion
2. 2Use self-reconstruction to further refine the generated pseudo-fake samples
3. 3Train audio deepfake detectors on the generated pseudo-fake samples to mitigate bias
4. 4Evaluate the performance of the detectors on unseen datasets to assess the effectiveness of AFSS

AI engineers and researchers working on audio deepfake detection can benefit from AFSS to improve the generalization of their models across unseen datasets. This can also be useful for data scientists working on fairness and bias mitigation in machine learning models

💡 Generating pseudo-fake samples from real audio can help mitigate bias in audio deepfake detection

🔊 Mitigate bias in audio deepfake detection with AFSS! 🤖