jazznet Dataset for Music Audio Machine Learning Research

The jazznet dataset is a dataset containing 162520 labeled piano patterns: chords, arpeggios, scales, and chord progressions, and their inversions in all keys of the 88-key piano. This results in ~95GB and more than 26K hours of audio. The patterns are guided by the jazz piano genre, but encompass other genres, like country, pop, blues, etc. The project GitHub page contains details on how to download the dataset or easily generate new data. You may also download the dataset directly from Zenodo.

Audio samples (all samples are in the middle C and in the root form)

C4maj7 chord


C4maj7 arpeggio


C4-ii-V-I-maj


C4-ii-triV-I


C4-I-VI-ii-V


C4-I-i#-ii-V


C4-iii-VI-ii-V


C4-ii#-V#-ii-V


C4-mixolydian


C4-pentatonic