MTG-Jamendo: Music Autotagging Metadata with Genre and Mood Labels

MTG-Jamendo provides metadata, scripts, and baselines for music autotagging research, created by the Music Technology Group (MTG). It serves as a benchmark for audio analysis tasks using tracks sourced from the Jamendo platform under Creative Commons licenses.

Use Cases

• Training multi-label classification models using genre, instrument, and mood tags
• Evaluating audio representation learning using the provided baseline scripts
• Benchmarking music information retrieval (MIR) systems against standardized data splits

Strengths

• Apache-2.0 licensed metadata and scripts
• Multi-label annotations covering genre, instrument, and mood
• Standardized evaluation splits for reproducible benchmarking

Limitations

• Potential label noise due to user-contributed tags from the Jamendo platform
• Requires significant storage and bandwidth to download the associated audio files

Provenance

Source

Music Technology Group (MTG), Universitat Pompeu Fabra

Collection Method

Aggregated from Jamendo platform metadata and user tags

Freshness

Last updated January 2026.

Geography

Global