dada-bots / dadaGP

Encoder/decoder converts GuitarPro songs to/from a token-sequence format for generative language models like GPT2, TransformerXL, etc.

DadaGP

Paper | Generation Results | ISMIR Poster

DadaGP is:

• a dataset of 26,181 GuitarPro songs in 739 genres, converted to a token sequence format suitable for generative language models like GPT2, TransformerXL, etc.
• an encoder/decoder (v1.1) that converts gp3, gp4, gp5 files to/from this token format.

Please contact Dadabots or Pedro Sarmento via email or twitter, @dadabots / @umpedronosapato, to request access to the dataset for research purposes.

Usage

Requirements

• python3
• PyGuitarPro 0.6 (it ONLY works with 0.6 -- if you're using a newer version, install 0.6 in a virtual environment to run dadagp.py)

pip install 'PyGuitarPro==0.6'

ENCODE (guitar pro --> tokens)

python dadagp.py encode input.gp3 output.txt [artist_name]
python dadagp.py encode examples/progmetal.gp3 progmetal.tokens.txt unknown

DECODE (tokens --> guitar pro)

python dadagp.py decode input.txt output.gp5
python dadagp.py decode progmetal.tokens.txt progmetal.decoded.gp5

Note:

• only gp3, gp4, gp5 files are supported by the encoder;
• rare combinations of instruments and tunings may not be supported;
• banjo is not supported;
• instrument-change events are not supported;

How to Cite

@inproceedings{dadagp2021,
  author = {Sarmento, Pedro and Kumar, Adarsh and Carr, CJ and Zukowski, Zack and Barthet, Mathieu and Yang, Yi-Hsuan},
  booktitle = {Proceedings of the 22nd International Society for Music Information Retrieval Conference},
  title = {{DadaGP: a Dataset of Tokenized GuitarPro Songs for Sequence Models}},
  url = {https://archives.ismir.net/ismir2021/paper/000076.pdf},
  year = {2021}
}