BROWSE
Detail
Conference Paper
How Far Can We Go With Synthetic Data for Audio-Visual Sound Source Localization?
Cited - time in 
Share 
Share
- Authors
- Arda Senocak, Sooyoung Park, Tae-Hyun Oh, Joon Chung
- Issue Date
- 2026-06
- Citation
- Conference on Computer Vision and Pattern Recognition (CVPR) 2026, pp.22931-22940
- Publisher
- Computer Vision Foundation
- Language
- English
- Type
- Conference Paper
- Abstract
- We present the first scalable framework for training sound source localization (SSL) models using synthetic data from text-to-X models. Although SSL has made notable progress, existing models remain constrained by limited-scale, un- curated real-world datasets that often suffer from seman- tic misalignment. Furthermore, the introduction of new SSL tasks and benchmarks has increased the need for more generalizable models. To address these challenges, we leverage synthetic data to create synthetic clones of the VGGSound dataset, enabling both fully synthetic and hy- brid real–synthetic training. We demonstrate that synthetic data can effectively replace, refine, and scale real train- ing datasets. Extensive experiments across multiple bench- marks show that synthetic data not only matches real data in performance but also enables significant improvements when combined with real samples. Our findings provide the first systematic evidence that synthetic data can serve as a scalable and effective approach for advancing SSL mod- els. Code and data are available at: https://github. com/swimmiing/SyntheticSSL.
- KSP Keywords
- Audio-visual, Real data, Real samples, Real-world, Scalable framework, Synthetic data, need for, sound source localization