Research map / Image generation / Fast diffusion sampling
Token-Based Audio Inpainting via Discrete Diffusion
Tali Dror, Iftach Shoham, Moshe Buchris, Oren Gal, Haim H. Permuter, Gilad Katz, Eliya Nachmani
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 span-based masking (ablation) | AIDD base · MusicNet (300 ms gaps) | FAD: 4.015 -> 3.573 | Improved |
| add derivative-based loss (lambda=200, first-order) (ablation) | AIDD base · MusicNet (300 ms gaps) | FAD: 4.015 -> 3.439 | Improved |
| raise derivative loss weight from 200 to 800 (first-order) (ablation) | AIDD · MusicNet (300 ms gaps) | FAD: 3.439 -> 3.837 | Hurt |
| combine span masking and derivative loss (ablation) | AIDD · MusicNet (200 ms gaps) | FAD: 2.802 -> 2.391 | Improved |
| combine span masking and derivative loss vs derivative alone (ablation) | AIDD · MusicNet (300 ms gaps) | FAD: 3.439 -> 3.549 | No clear effect |
| use WavTokenizer instead of UniCodec (ablation) | AIDD · MAESTRO (750 ms gap) | ODG (PEA-Q): -3.289 -> -2.596 | Improved |
| use AIDD instead of CQT-Diff+ | AIDD · MusicNet (150 ms gaps) | FAD: 1.525 -> 1.866 | Hurt |
| use AIDD instead of CQT-Diff+ | AIDD · MusicNet (300 ms gaps) | FAD: 4.652 -> 3.549 | Improved |
| use AIDD instead of bin2bin-MIDI | AIDD · MAESTRO (375 ms gap) | ODG (PEA-Q): -2.800 -> -2.303 | Improved |
| use AIDD instead of CQT-Diff+ (subjective) | AIDD · MAESTRO (subjective) | MOS: 3.51 -> 3.64 | Improved |
Closest papers by meaning
- DITTO: Diffusion Inference-Time T-Optimization for Music Generation (ICML 2024 · similarity 0.87)
- WavTokenizer: an Efficient Acoustic Discrete Codec Tokenizer for Audio Language Modeling (ICLR 2025 · similarity 0.87)
- Fast Timing-Conditioned Latent Audio Diffusion (ICML 2024 · similarity 0.86)
- ZeroSep: Separate Anything in Audio with Zero Training (NeurIPS 2025 · similarity 0.86)
- Live Music Diffusion Models: Efficient Fine-Tuning and Post-Training of Interactive Diffusion Music Generators (NeurIPS 2026 · similarity 0.86)
- Prompt-guided Precise Audio Editing with Diffusion Models (ICML 2024 · similarity 0.86)
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.
