Freesound Dataset 50k Human-Labeled Audio Events

FSD50K is an open dataset of 50,000 human-labeled sound events. It was created by researchers including Eduardo Fonseca and Xavier Serra for audio classification tasks.

Use Cases

• Train audio classification models using the 50,000 labeled sound events.
• Analyze patterns in human-labeled audio events across different categories.
• Benchmark audio event detection systems on a large-scale labeled dataset.

Strengths

• Contains 50,000 labeled audio samples.
• Dataset is publicly available and cited in a 2020 arXiv paper.

Limitations

• Specific column structure, audio features, and label categories are not detailed in the provided input.
• Geographic and temporal coverage of the audio samples is unspecified.

Provenance

Source

Original dataset located at Zenodo, created by researchers including Eduardo Fonseca and Xavier Serra.

Collection Method

Human-labeled sound events, methodology detailed in the associated arXiv paper.

Freshness

Dataset copy was last updated on the platform in May 2025.

Geography

Region tag indicates 'us', but specific spatial coverage is unknown.