Data from: Performer and Subgenre Conditioned Generation of Jazz Piano Music

This data accompanies our paper "Performer and Subgenre Conditioned Generation of Jazz Piano Music". For more information, see our code repository. Downloading This metadata is provided for academic research purposes only and the material contained within it should not be used for any commercial purpose under the terms of the CC BY-NC-SA license. Access will only be granted for research projects, and potential users of the data must apply for access. These requests are checked manually: please do not fill in the form multiple times, we will aim to grant you access as soon as possible. After you have been granted access, you will be able to download a single `.zip` file that has the same structure as our code repository. The `.zip` file contains all of the MIDI files and metadata used to train our model, as well as checkpoints for the pre-trained and fine-tuned model, and the model trained with reinforcement learning. You should first clone our repository, then extract the zip file into the root directory. You should end up with the following file structure. .└── jazz-style-conditioned-generation/ ├── checkpoints/ │ ├── finetuning-customtok-plateau/ # pretrained on ATEPP │ │ └── finetuning_customtok_10msmin***/ │ │ ├── tokenizer.json # dumped tokenizer configuration │ │ └── validation_best.pth # pytorch model │ ├── pretraining-custom-tokenizer-fixed-preprocessing # finetuned on jazz │ │ └── pretraining_customtok_10msmin***/ │ └── reinforcement-customtok-plateau # finetuned + DPO-P ├── config/ # same file structure as `checkpoints/` │ ├── finetuning-customtok-plateau/ # individual folders per experiment │ │ └── finetuning_customtok_10msmin***.yaml # individual .yaml files per run │ ├── pretraining-custom-tokenizer-fixed-preprocessing │ │ └── ***.yaml │ └── reinforcement-customtok-plateau │ │ └── ***.yaml ├── data/ │ ├── pretraining/ │ │ └── atepp/ │ │ ├── one_folder_per_recording/ │ │ └── ... │ └── raw/ # one folder per source dataset │ ├── bushgrafts │ ├── jja │ ├── jtd/ │ │ ├── one_folder_per_recording/ # all recordings have the same directory structure │ │ │ ├── piano_midi.mid # raw MIDI data │ │ │ └── metadata_tivo.json # scraped TiVo metadata │ │ └── ... │ ├── pijama │ └── pianist8 └── README.md # --> you are here! For more information on how to use this data to reproduce our training results or run inference, see the README.