Research map / Video / Video segmentation and tracking
Into the Wild with AudioScope: Unsupervised Audio-Visual Separation of On-Screen Sounds
Efthymios Tzinis, Scott Wisdom, Aren Jansen, Shawn Hershey, Tal Remez, Dan Ellis, 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 |
|---|---|---|---|
| Add a small amount of human-labeled training examples | AudioScope · YFCC100m mixture of mixtures | OSR (dB): 5.3 dB and 10.7 dB | Improved |
| Use exact cross-entropy instead of active combinations or multiple-instance loss | AudioScope · YFCC100m mixture of mixtures | OSR (dB): 10.8; 9.4; 5.3 | Hurt |
| Add 25% synthetic off-screen examples to semi-supervised training | AudioScope · YFCC100m mixture of mixtures | OSR (dB): 10.7 -> 61.5 | Improved |
| Remove video conditioning from separation (ablation) | AudioScope · YFCC100m single mixtures | SI-SNR, on-screen estimate: 13.5 -> 11.5 | Hurt |
| Remove video conditioning from separation (ablation) | AudioScope · YFCC100m mixture of mixtures | SI-SNR, on-screen estimate: 6.3 -> 5.4 | Hurt |
| Train audio and video embedding networks from scratch instead of using pretrained weights (ablation) | AudioScope · YFCC100m single mixtures | AUC: 0.58 -> 0.45 | Hurt |
| Remove global video and audio embeddings from classifier input (ablation) | AudioScope, unsupervised · YFCC100m single mixtures | SI-SNR, on-screen estimate: 13.5 -> 14.0 | Mixed |
| Reduce attention embedding dimension from 256 to 128 (ablation) | AudioScope, unsupervised · YFCC100m single mixtures | SI-SNR, on-screen estimate: 13.5 -> 12.1 | Hurt |
| Replace attentional pooling for video with mean pooling (ablation) | AudioScope, unsupervised · YFCC100m mixture of mixtures | OSR: 9.4 -> 6.2 | Hurt |
| Use unfiltered on-screen and filtered off-screen data for unsupervised training (ablation) | AudioScope · YFCC100m single mixtures | AUC: 0.58 -> 0.63 | Improved |
| Use unfiltered on-screen and filtered off-screen data for unsupervised training (ablation) | AudioScope · YFCC100m mixture of mixtures | OSR: 9.4 -> 5.6 | Hurt |
| Increase output sources from 4 to 8 (ablation) | AudioScope, unsupervised · YFCC100m mixture of mixtures | SI-SNR, MixIT*: 10.5 dB -> 11.1 dB | Improved |
Closest papers by meaning
- Continual Audio-Visual Sound Separation (NeurIPS 2024 · similarity 0.88)
- Seeing Speech and Sound: Distinguishing and Locating Audio Sources in Visual Scenes (CVPR 2025 · similarity 0.88)
- Dynamic Derivation and Elimination: Audio Visual Segmentation with Enhanced Audio Semantics (CVPR 2025 · similarity 0.88)
- Unraveling Instance Associations: A Closer Look for Audio-Visual Segmentation (CVPR 2024 · similarity 0.87)
- From Vision to Audio and Beyond: A Unified Model for Audio-Visual Representation and Generation (ICML 2024 · similarity 0.87)
- Weakly-supervised Audio Separation via Bi-modal Semantic Similarity (ICLR 2024 · 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.
