History of Recorded Jazz: DTL1000, 1920-2020

We present the DTL1000 dataset, which was created in the “Dig That Lick” project and covers the history of recorded jazz with a sample of 1,750 improvisations extracted from 1,060 audio tracks. The dataset contains a mixture of collected (editorial metadata), manually annotated (structure, style), and automatically generated (main melody transcriptions of solos) data describing the recordings. The motivation for creating this dataset was the study of patterns in jazz improvisation, but there are many other applications for this resource. The accompanying paper presents the dataset creation process, data structure and contents with descriptive statistics and discusses the origin and process of the annotations, as well as general use cases and specifically the case of pattern analysis. These components and their combinations enable a number of use cases for jazz studies as well as algorithm development for music analysis. The DTL1000 dataset provides a rich resource for a variety of disciplines, and constitutes a contribution to a field where large datasets with rich annotations are scarce.