Skip to content
Fan Zeng (Frank)
Back to resources

FSD50K

FSD50K provides human-labeled sound events from Freesound, with clips that may carry several event labels.

Visit official source (opens in a new tab)

Research use

Measure multi-label audio tagging using the published development and evaluation splits.

Limitations

Examples can contain overlapping foreground and background events. Clip licensing varies across the collection.

License and access

Dataset license
Per-clip Creative Commons licenses (opens in a new tab)
Access status
Direct access

Follow the instructions for extracting the split audio archives. Clip-level license and attribution metadata accompany the release.

Additional sources