Yumeng Zhang, Junyi Liu, Yifan Zhou
We use GiantMIDI-Piano and Pop909 as our training data. You can also access the dataset from the following Google Drive link:
GiantMIDI-Piano: https://drive.google.com/drive/folders/1IfI70HxlBKiuUGPT4_0OY4bZG0pmJz_x?usp=sharing
Pop909: https://drive.google.com/drive/folders/1KSwmvu4SGJM2arzWiNDE3h1UEwk7koPI?usp=sharing
GPT-2: To run the training and generation code, clone the repository
git clone https://github.com/EEexplorer001/symbolic_music_generation_transformers.git
Then run the run.ipynb in the GPT_music folder. You can change the data path, token path, model save path and generation path to your own.
Theme transformer: run ThemeTransformer.ipynb in theme_transformer folder.
Some of the code is based on the following two code base:
Miditok: https://github.com/Natooz/MidiTok
Theme transformer: https://github.com/atosystem/ThemeTransformer