DESED_public_eval: Domestic Environment Sound Event Detection Clips

DESED_public_eval is a collection of real 10-second audio clips for evaluating sound event detection systems. The clips are sourced from YouTube under Creative Commons licenses and curated by Nicolas Turpault of the Centre National de la Recherche Scientifique. This dataset serves as the public evaluation set, referred to as the 'youtube' evaluation in associated research papers.

Use Cases

• Benchmarking sound event detection models based on real-world audio clips.
• Training audio classifiers for domestic environments based on YouTube-sourced sounds.
• Evaluating model generalization on unseen, in-the-wild audio data.
• Researching acoustic scene analysis using Creative Commons licensed content.

Strengths

• Clips are 10 seconds long, providing a standardized duration for evaluation.
• Audio is sourced from real YouTube videos under Creative Commons licenses.
• Dataset is explicitly designed as a public evaluation benchmark for the research community.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

YouTube videos with Creative Commons licenses.

Collection Method

Clips were likely extracted and annotated for sound events.