A Data-Driven Diffusion-based Approach for Audio Deepfake Explanations
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914167818027008 |
|---|---|
| author | Grinberg, Petr Kumar, Ankur Koppisetti, Surya Bharaj, Gaurav |
| author_facet | Grinberg, Petr Kumar, Ankur Koppisetti, Surya Bharaj, Gaurav |
| contents | Evaluating explainability techniques, such as SHAP and LRP, in the context of audio deepfake detection is challenging due to lack of clear ground truth annotations. In the cases when we are able to obtain the ground truth, we find that these methods struggle to provide accurate explanations. In this work, we propose a novel data-driven approach to identify artifact regions in deepfake audio. We consider paired real and vocoded audio, and use the difference in time-frequency representation as the ground-truth explanation. The difference signal then serves as a supervision to train a diffusion model to expose the deepfake artifacts in a given vocoded audio. Experimental results on the VocV4 and LibriSeVoc datasets demonstrate that our method outperforms traditional explainability techniques, both qualitatively and quantitatively. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_03425 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | A Data-Driven Diffusion-based Approach for Audio Deepfake Explanations Grinberg, Petr Kumar, Ankur Koppisetti, Surya Bharaj, Gaurav Audio and Speech Processing Artificial Intelligence Machine Learning Evaluating explainability techniques, such as SHAP and LRP, in the context of audio deepfake detection is challenging due to lack of clear ground truth annotations. In the cases when we are able to obtain the ground truth, we find that these methods struggle to provide accurate explanations. In this work, we propose a novel data-driven approach to identify artifact regions in deepfake audio. We consider paired real and vocoded audio, and use the difference in time-frequency representation as the ground-truth explanation. The difference signal then serves as a supervision to train a diffusion model to expose the deepfake artifacts in a given vocoded audio. Experimental results on the VocV4 and LibriSeVoc datasets demonstrate that our method outperforms traditional explainability techniques, both qualitatively and quantitatively. |
| title | A Data-Driven Diffusion-based Approach for Audio Deepfake Explanations |
| topic | Audio and Speech Processing Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2506.03425 |