mlsanityData to 14 Oct 2026

The Perceived Fragility of Explanations in Audio Models: Manipulation of Attribution with Unchanged Predictions

arXiv 2606.14466 ↗First posted 12 Jun 2026cs.SD cs.AI cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

This paper investigates the fragility of post-hoc explanation methods in audio deepfake detection. While previous work on explanation manipulation focused on images using standard Lp metrics, we introduce a psychoacoustic framework that optimizes inaudible perturbations to decouple model attributions from final classifications. We evaluate this vulnerability across state-of-the-art architectures under strict prediction-preserving constraints. By evaluating the manipulation cost through domain-specific perceptual audio quality metrics alongside explanation alignment criteria, our framework demonstrates that an adversary can systematically distort automated explanation heatmaps while preserving the predicted deepfake label. Full code available at: https://github.com/cncPomper/Audio-XAI

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 12 of its 19 references that are in the map. First posted on arXiv 12 Jun 2026.

Builds on

12 of 19 references.

References · 12

  1. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  2. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,415
  3. 2016CNN Architectures for Large-Scale Audio Classificationcited by 704
  4. 2017Interpretation of Neural Networks is Fragilecited by 444
  5. 2019Explanations can be manipulated and geometry is to blamecited by 216
  6. 2019Fooling Neural Network Interpretations via Adversarial Model Manipulationcited by 112
  7. 2021CDPAM: Contrastive learning for perceptual audio similaritycited by 42
  8. 2024SONICS: Synthetic Or Not -- Identifying Counterfeit Songscited by 21
  9. 2025Zimtohrli: An Efficient Psychoacoustic Audio Similarity Metriccited by 6
  10. 2021Beyond Lp clipping: Equalization-based Psychoacoustic Attacks against ASRscited by 4
  11. 2025A Data-Driven Diffusion-based Approach for Audio Deepfake Explanationscited by 2
  12. 2026Right Predictions, Misleading Explanations: On the Vulnerability of Vision-Language Model Explanationscited by 1

Built on by

Not cited yet.