JazzSAMBA: A Synchronous and Asynchronous Multi-take Band Audio Dataset of Jazz Standards for Live Music Models

JazzSAMBA (Jazz Synchronous and Asynchronous Multi-take Band Audio) is a multitrack dataset of 76 jazz standards recorded by eight musicians (drums, bass, piano, trumpet, saxophone). Songs use either an asynchronous per-instrument protocol or a synchronous full-band protocol, with preferred/alternate take tiers, stem audio, mixtures, MIDI, and timed bar-grid annotations (bars, chords, sections, soloists). Project page (listening examples, explore, paper/code links): https://pnlong.github.io/jazzsamba-demo/ Python loader: pip install jazzsamba (point it at a local JazzSAMBA/ folder after unzipping). Files (one record) Zip Contents JazzSAMBA-catalog.zip README.md, LICENSE, songs.csv, musicians.csv, debleed.json JazzSAMBA-async.zip Asynchronous recorded stems / mixtures / annotations / splits JazzSAMBA-sync.zip Synchronous recorded stems / mixtures / annotations / splits JazzSAMBA-async-derived.zip Async derived overlays (AMT MIDI, etc.) JazzSAMBA-sync-derived.zip Sync derived overlays (AMT MIDI, debleeded audio, etc.) Zip members are relative to the JazzSAMBA root. See the catalog README.md for layout and citation.