Research map / Video / Video generation
Physics-Driven Diffusion Models for Impact Sound Synthesis From Videos
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| condition the diffusion model on physics priors instead of spectrogram features | Video-conditioned diffusion for impact sound synthesis · Greatest Hits test | FID: 28.77 -> 26.20 | Helped |
| condition the diffusion model on physics priors instead of spectrogram features | Video-conditioned diffusion for impact sound synthesis · Greatest Hits test | KID (mean, std): 0.016, 0.009 -> 0.010, 0.008 | Helped |
| condition the diffusion model on physics priors instead of class labels | Video-conditioned diffusion for impact sound synthesis · Greatest Hits test | KL Divergence: 2.38 -> 2.04 | Helped |
| condition the diffusion model on physics priors instead of class labels | Video-conditioned diffusion for impact sound synthesis · Greatest Hits test | Recognition accuracy: 72.02 -> 74.09 | Helped |
| use the physics-driven diffusion method rather than the transformer-based model | Impact sound synthesis · Greatest Hits human survey | Matching selection: 26.6% -> 55.4% | Helped |
| use the physics-driven diffusion method rather than the transformer-based model | Impact sound synthesis · Greatest Hits human survey | Quality selection: 28.8% -> 53.6% | Helped |
| add physics priors (ablation) | Video-only impact sound diffusion · Greatest Hits human survey | Matching selection: 23.6% -> 38.6% | Helped |
| add physics priors (ablation) | Video-only impact sound diffusion · Greatest Hits human survey | Quality selection: 23.6% -> 40.6% | Helped |
| replace class labels with physics priors (ablation) | Video + label impact sound diffusion · Greatest Hits human survey | Matching selection: 37.8% -> 38.6% | Helped |
| replace class labels with physics priors (ablation) | Video + label impact sound diffusion · Greatest Hits human survey | Quality selection: 35.8% -> 40.6% | Helped |
| use the proposed diffusion method rather than retrieval from predicted sound features | Impact sound synthesis · Greatest Hits retrieval comparison | KL Divergence: 7.39 -> 2.04 | Helped |
| use the proposed diffusion method rather than retrieval from predicted sound features | Impact sound synthesis · Greatest Hits retrieval comparison | Spectrogram MSE: 0.205 -> 0.149 | Helped |
Closest papers by meaning
- AV-GS: Learning Material and Geometry Aware Priors for Novel View Acoustic Synthesis (NeurIPS 2024 · similarity 0.85)
- SonicSim: A customizable simulation platform for speech processing in moving sound source scenarios (ICLR 2025 · similarity 0.85)
- Resounding Acoustic Fields with Reciprocity (NeurIPS 2025 · similarity 0.85)
- Acoustic Volume Rendering for Neural Impulse Response Fields (NeurIPS 2024 · similarity 0.85)
- Read, Watch and Scream! Sound Generation from Text and Video (AAAI 2025 · similarity 0.84)
- SoundCTM: Unifying Score-based and Consistency Models for Full-band Text-to-Sound Generation (ICLR 2025 · similarity 0.84)
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.