JazzSAMBA

A multitrack dataset of jazz standards recorded both asynchronously (per-instrument sessions) and synchronously (full band together), with timed bar, chord, section, and soloist annotations for live music models.

76songs
50asynchronous
26synchronous
8musicians

Why JazzSAMBA

Interactive accompaniment and multitrack MIR need realistic combo audio with reliable structure labels. JazzSAMBA pairs original studio performances with a shared bar grid, better/worse take tiers, and public stem FLACs on Zenodo.


Dataset at a glance

Asynchronous (per-instrument) and synchronous (full-band) takes, stem and mixture assets, and form-aligned annotations on a shared bar grid.

B and W denote better and worse take tiers. Asynchronous songs include trumpet and one saxophone, either alto or tenor but not both. Drums are recorded first; later instruments may follow in any order, and the sequence shown is only one example. Annotations authored on the better mixture are copied onto the worse tier. Synchronous songs include trumpet and/or saxophone with the full band recording together, and each take is annotated separately. Click the figure for full screen.

Featured mixes

One-minute clips of better-take mixes (24 kHz / 192 kbps), with stems under each. Explore plays full better-take mixtures only; stems and worse takes ship in the dataset download.


Cite

@inproceedings{jazzsamba, title={JazzSAMBA: A Synchronous and Asynchronous Multi-take Band Audio Dataset of Jazz Standards for Live Music Models}, author={Long, Phillip and Nguyen, Jacob and Hosto, Jace and Hosto, Gage and Takazawa, Jett and Nofal, Fares and Stade, Sebastian and Shikarpur, Nithya and McAuley, Julian and Huang, Cheng-Zhi Anna and Ma, Aleksandra Teng and Brade, Stephen}, note={DOI and venue forthcoming}, year={2026} }

Zenodo DOI and camera-ready BibTeX will replace this placeholder.