Research map / Video / Video segmentation and tracking
Self-Supervised Audio-Visual Soundscape Stylization
Tingle Li, Renhao Wang, Po-Yao Huang, Andrew Owens, Gopala Krishna Anumanchipalli
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 Gaussian noise to the enhanced input audio (ablation) | audio-visual soundscape stylization model · CityWalk | FAD: 2.66 -> 1.64 | Improved |
| condition on speech-containing audio instead of only separated sound effects (ablation) | audio-visual soundscape stylization model · CityWalk | RTE: 0.65 -> 0.20 | Improved |
| use two-stage enhancement (separation then enhancement) instead of separation only (ablation) | audio-visual soundscape stylization model · CityWalk | RTE: 0.60 -> 0.20 | Improved |
| use two-stage enhancement instead of enhancement-only or separation-only (ablation) | enhancement strategy · CityWalk | OVL: 3.221 -> 3.987 | Improved |
| keep matched audio-visual conditioning rather than randomly swapped audio or image (ablation) | audio-visual soundscape stylization model · CityWalk | FAD: 4.92 / 4.84 -> 1.64 | Improved |
| add visual conditioning to fine-tuned audio conditioning (ablation) | audio-visual soundscape stylization model · CityWalk | FAD: 2.08 -> 1.64 | Improved |
| add classifier-free guidance (ablation) | audio-visual soundscape stylization model · CityWalk | FAD: 3.37 -> 1.64 | Improved |
| fine-tune the CLAP audio encoder (ablation) | audio conditioning · CityWalk | FAD: 6.12 -> 2.08 | Improved |
| fine-tune the CLIP image encoder (ablation) | visual conditioning · CityWalk | FAD: 2.28 -> 2.29 | No clear effect |
| use pretrained CLIP and CLAP encoders instead of a ResNet-18 trained from scratch (ablation) | audio-visual soundscape stylization model · CityWalk | FAD: 2.41 -> 1.64 | Improved |
| do not normalize loudness of inputs (ablation) | audio-visual soundscape stylization model · CityWalk | FAD: 2.54 -> 1.64 | Improved |
| extend audio conditioning from 2.56s to 5.12s (ablation) | audio-only conditioning · CityWalk | FAD: 2.08 -> 2.10 | No clear effect |
Closest papers by meaning
- Self-Supervised Learning of Audio-Visual Objects from Video (ECCV 2020 · similarity 0.90)
- Visual Acoustic Matching (CVPR 2022 · similarity 0.89)
- Audio-Visual Contrastive Learning with Temporal Self-Supervision (AAAI 2023 · similarity 0.89)
- SoundBrush: Sound as a Brush for Visual Scene Editing (AAAI 2025 · similarity 0.87)
- ES3: Evolving Self-Supervised Learning of Robust Audio-Visual Speech Representations (CVPR 2024 · similarity 0.87)
- Seeing Speech and Sound: Distinguishing and Locating Audio Sources in Visual Scenes (CVPR 2025 · 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.
