Music datasets
Find training and evaluation resources, with separate notes on recordings, metadata and permitted uses.
| Dataset | Format | Use |
|---|---|---|
| GTZAN | Genre-labeled music excerpts | A common historical genre-classification benchmark, useful for reproducing and examining older MIR results. |
| MAESTRO v3 | Piano audio + MIDI | Piano transcription, expressive timing, velocity and pedal analysis. |
| NSynth | Isolated notes + labels | Timbre, pitch and instrument-family analysis; neural synthesis experiments. |
| Groove MIDI Dataset | Drum MIDI + audio | Microtiming, velocity, groove and drummer-performance analysis. |
| MUSDB18 / MUSDB18-HQ | Mixes + stems | The widely used four-source music-separation benchmark: vocals, drums, bass and other. |
| Free Music Archive (FMA) | Audio + metadata + features | Genre recognition, retrieval, representation learning and music analysis. |
| MTG-Jamendo | Music audio + tags | Genre, instrument and mood/theme tagging with published benchmark splits. |
| Lakh MIDI Dataset | MIDI + optional alignment | Large-scale symbolic analysis, note statistics, alignment and sequence modeling. |
| Slakh2100 | Synthesized stems + MIDI | Instrument transcription and separation with aligned symbolic ground truth. |
| GuitarSet | Guitar audio + annotations | Pitch, note, string/fret, chord, beat and playing-style analysis. |
| MusicCaps | Captions + YouTube segment IDs | Music captioning, text/audio retrieval and evaluation of text-conditioned generation. |
| AudioSet | Sound labels + video IDs | General sound-event detection, including musical instruments and music categories. |
| MedleyDB | Multitracks + melody annotations | Melody estimation, instrument activity and source-separation research. |
| MagnaTagATune | Music clips + listener tags | An established music-tagging benchmark for instrument, genre and mood labels. |