Research map / Vision-language and multimodal / Cross-modal learning
AudioScopeV2: Audio-Visual Attention Architectures for Calibrated Open-Domain On-Screen Sound Separation
Efthymios Tzinis, Scott Wisdom, Tal Remez, John R. Hershey
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 |
|---|---|---|---|
| use joint cross-modal attention instead of AudioScope shallow attention, trained on matched unfiltered data | AudioScope* · unfiltered random background test | SNR (dB): 5.8 -> 7.7 | Improved |
| raise frame rate from 1 to 16 FPS | Joint CMA · unfiltered random background test | SNR (dB): 7.3 -> 7.7 | Improved |
| raise frame rate from 1 to 16 FPS | AudioScope* · unfiltered random background test | SNR (dB): 5.8 -> 5.2 | Hurt |
| use joint instead of separable self-attention | SA model · unfiltered random background test (1 FPS) | SNR (dB): 6.6 -> 7.2 | Improved |
| use joint instead of separable cross-modal attention | CMA model · unfiltered random background test (1 FPS) | SNR (dB): 6.4 -> 7.3 | Improved |
| train on unfiltered data with separation pretraining (AudioScope*) | AudioScope · filtered test (1 FPS) | SNR (dB): 6.0 -> 8.2 | Improved |
| halve input audio as on-screen estimate (6 dB OSR) | no-processing baseline · unfiltered random background test | SNR (dB): 2.5 -> 4.1 | Improved |
| use 8 blocks instead of 4 (ablation) | Separable SA · unfiltered test, OSR 6 dB | SNR (dB): 6.6 -> 3.5 | Hurt |
| use 2 attention heads instead of 4 (ablation) | Separable SA · unfiltered test, OSR 6 dB | SNR (dB): 6.6 -> 7.4 | Improved |
| use 8 attention heads instead of 4 (ablation) | Separable SA · unfiltered test, OSR 6 dB | SNR (dB): 6.6 -> 6.2 | Hurt |
| remove dropout (ablation) | Separable SA · unfiltered test, OSR 6 dB | SNR (dB): 6.6 -> 7.0 | Improved |
| use 4x4 spatial locations instead of 8x8 (ablation) | Separable SA · unfiltered test, OSR 6 dB | SNR (dB): 6.6 -> 7.0 | Improved |
Closest papers by meaning
- Into the Wild with AudioScope: Unsupervised Audio-Visual Separation of On-Screen Sounds (ICLR 2021 · similarity 0.94)
- Continual Audio-Visual Sound Separation (NeurIPS 2024 · similarity 0.88)
- Learning to Separate Object Sounds by Watching Unlabeled Video (ECCV 2018 · similarity 0.88)
- AlignSep: Temporally-Aligned Video-Queried Sound Separation with Flow Matching (ICLR 2026 · similarity 0.88)
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.86)
- 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.
