Research map / Image generation / Autoregressive and latent generation
VoiceCloak: A Multi-Dimensional Defense Framework Against Unauthorized Diffusion-Based Voice Cloning
Qianyue Hu, Junyan Wu, Wei Lu, Xiangyang Luo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add L_ID (WavLM embedding distance with opposite-gender centroid) (ablation) | VoiceCloak · LibriTTS (DiffVC) | ASV acceptance rate (%): 46.82% -> 8.57% | Improved |
| add opposite-gender centroid guidance to L_ID (ablation) | VoiceCloak L_ID · LibriTTS (DiffVC) | ASV acceptance rate (%): 19.92% -> 8.57% | Improved |
| add opposite-gender centroid guidance to L_ID (ablation) | VoiceCloak L_ID · LibriTTS (DiffVC) | Defense success rate (%): 14.40% -> 27.74% | Improved |
| add attention context divergence L_ctx (ablation) | VoiceCloak · LibriTTS (DiffVC) | Defense success rate (%): 22.58% -> 62.57% | Improved |
| add L_ctx to L_ID (ablation) | VoiceCloak · LibriTTS (DiffVC) | Defense success rate (%): 27.74% -> 69.20% | Improved |
| add score magnitude amplification L_score (ablation) | VoiceCloak · LibriTTS (DiffVC) | Defense success rate (%): 20.20% -> 41.20% | Improved |
| add noise-guided semantic corruption L_sem (ablation) | VoiceCloak · LibriTTS (DiffVC) | Defense success rate (%): 20.20% -> 60.60% | Improved |
| add Sem-free (noise-feature) term to L_sem (ablation) | VoiceCloak L_sem · LibriTTS (DiffVC) | Defense success rate (%): 26.80% -> 60.60% | Improved |
| add L_score to L_sem (ablation) | VoiceCloak · LibriTTS (DiffVC) | Defense success rate (%): 60.60% -> 57.80% | Mixed |
| increase perturbation budget eps from 0.0005 to 0.0100 (ablation) | VoiceCloak · LibriTTS (DiffVC) | Defense success rate (%): 41.20% -> 90.80% | Improved |
| apply 32 kbps MP3 compression to protected audio | VoiceCloak · LibriTTS (DiffVC) | Defense success rate (%): 66.42% -> 29.40% | Hurt |
| apply VoiceCloak to unseen target DDDM-VC instead of DiffVC | VoiceCloak · LibriTTS | Defense success rate (%): 71.40% -> 54.89% | Hurt |
Closest papers by meaning
- De-AntiFake: Rethinking the Protective Perturbations Against Voice Cloning Attacks (ICML 2025 · similarity 0.91)
- RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (NeurIPS 2026 · similarity 0.88)
- Do Not Mimic My Voice : Speaker Identity Unlearning for Zero-Shot Text-to-Speech (ICML 2025 · similarity 0.87)
- E2E-VGuard: Adversarial Prevention for Production LLM-based End-To-End Speech Synthesis (NeurIPS 2025 · similarity 0.86)
- AudioMarkBench: Benchmarking Robustness of Audio Watermarking (NeurIPS 2024 · similarity 0.85)
- Silence is Golden: Leveraging Adversarial Examples to Nullify Audio Control in LDM-based Talking-Head Generation (CVPR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
