Research map / Video / Video generation
Step-by-Step Video-to-Audio Synthesis via Negative Audio Guidance
Akio Hayakawa, Masato Ishii, Takashi Shibuya, Yuki Mitsufuji
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 Negative Audio Guidance (NAG) | MMAudio-S-16k · Multi-Caps VGGSound individual tracks | CLAP A-A separability: 79.75 -> 71.38 | Improved |
| use negative prompting instead of NAG | MMAudio-S-16k · Multi-Caps VGGSound individual tracks | IS: 12.47 -> 11.19 | Hurt |
| add NAG to step-by-step generation vs negative prompting | MMAudio-S-16k · Multi-Caps VGGSound composite | FD_PANNs: 9.21 -> 6.47 | Improved |
| increase NAG coefficient beta from 0.0 to 2.0 (alpha 3.5) (ablation) | MMAudio-S-16k + NAG · Multi-Caps VGGSound individual tracks | CLAP A-A: 79.46 -> 71.97 | Improved |
| increase NAG coefficient beta from 0.0 to 2.0 (alpha 3.5) (ablation) | MMAudio-S-16k + NAG · Multi-Caps VGGSound individual tracks | IS: 12.27 -> 11.62 | Hurt |
| use descending (core event first) generation order instead of ascending (ablation) | MMAudio-S-16k + NAG · Multi-Caps VGGSound composite | FD_PANNs: 7.36 -> 6.47 | Improved |
| use descending order instead of random order (ablation) | MMAudio-S-16k + NAG · Multi-Caps VGGSound composite | FD_PANNs: 7.32 -> 6.47 | Improved |
| use step-by-step NAG instead of one-step generation plus AudioSep | MMAudio-S-16k · Multi-Caps VGGSound individual tracks | CLAP T-A: 23.42 -> 28.91 | Improved |
| use V2A NAG instead of independent T2A generation (AudioX) | AudioX · Multi-Caps VGGSound composite | IB-score: 19.54 -> 28.65 | Improved |
| apply NAG to ControlFoley | ControlFoley · ControlFoley individual tracks | CLAP A-A: 60.67 -> 53.81 | Improved |
| apply NAG to MMAudio | MMAudio-S-16k · AudioCaps test | CLAP A-A: 68.89 -> 60.43 | Improved |
| apply NAG to MMAudio | MMAudio-S-16k · Movie Gen Audio Bench | CLAP A-A: 72.25 -> 66.32 | Improved |
Closest papers by meaning
- Tri-Ergon: Fine-Grained Video-to-Audio Generation with Multi-Modal Conditions and LUFS Control (AAAI 2025 · similarity 0.89)
- V2Meow: Meowing to the Visual Beat via Video-to-Music Generation (AAAI 2024 · similarity 0.89)
- Read, Watch and Scream! Sound Generation from Text and Video (AAAI 2025 · similarity 0.88)
- V2A-Mapper: A Lightweight Solution for Vision-to-Audio Generation by Connecting Foundation Models (AAAI 2024 · similarity 0.88)
- VinTAGe: Joint Video and Text Conditioning for Holistic Audio Generation (CVPR 2025 · similarity 0.88)
- Frieren: Efficient Video-to-Audio Generation Network with Rectified Flow Matching (NeurIPS 2024 · similarity 0.88)
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.
