Research map / Vision-language and multimodal / Cross-modal learning
Cinematic Audio Source Separation Using Visual Cues
Kang Zhang, Suyeon Lee, Arda Senocak, Joon Son Chung
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 facial video stream to audio-only model (ablation) | AV-CASS · AVDnR | FAD: 1.63 -> 0.91 | Improved |
| add scene video stream to audio-only model (ablation) | AV-CASS · AVDnR | FAD: 1.63 -> 0.87 | Improved |
| add both facial and scene streams to audio-only model (ablation) | AV-CASS · AVDnR | FAD: 1.63 -> 0.84 | Improved |
| add both visual streams (ablation) | AV-CASS · AVDnR | KL: 1.15 -> 0.93 | Improved |
| add both visual streams (ablation) | AV-CASS · AVDnR | PESQ: 2.08 -> 2.26 | Improved |
| add both visual streams (ablation) | AV-CASS · AVDnR | SI-SDRi (dB): 12.23 -> 12.32 | No clear effect |
| add both visual streams (ablation) | AV-CASS · AVDnR | WPR (%): 2.01 -> 1.84 | Improved |
| add facial stream (ablation) | AV-CASS · AVDnR DX stem | WPR (%): 0.0265 -> 0.0164 | Improved |
| add scene stream (ablation) | AV-CASS · AVDnR FX stem | WPR (%): 4.6507 -> 4.0939 | Improved |
| add scene stream only (ablation) | AV-CASS · AVDnR DX stem | WPR (%): 0.0265 -> 0.0289 | Hurt |
| increase sampling steps to 32 (ablation) | AV-CASS · AVDnR | FAD | Improved |
| use dual-stream multi-source visual conditioning instead of single-target DAVIS-Flow | AV-CASS · AVDnR | FAD: 5.94 -> 0.84 | Improved |
Closest papers by meaning
- Into the Wild with AudioScope: Unsupervised Audio-Visual Separation of On-Screen Sounds (ICLR 2021 · similarity 0.87)
- Audio-Sync Video Generation with Multi-Stream Temporal Control (NeurIPS 2025 · similarity 0.86)
- Read, Watch and Scream! Sound Generation from Text and Video (AAAI 2025 · similarity 0.85)
- Seeing Speech and Sound: Distinguishing and Locating Audio Sources in Visual Scenes (CVPR 2025 · similarity 0.85)
- Learning to Highlight Audio by Watching Movies (CVPR 2025 · similarity 0.84)
- MMAudio: Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis (CVPR 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.
