RHEPP-Transformer: Reconstructing Human Expressiveness in Piano Performances with a Transformer Network

This repo presents the code implementation for the paper Reconstructing Human Expressiveness in Piano Performances with a Transformer Network

Training

The training was monitored by with W&B. The pre-trained model could be found and downloaded here.

For re-training the model, please contact me for the data and run the following commands:

python main.py --cuda_devices YOUR_CUDA_DEVICES

Generation

For generate expressive piano performance from transcribed score (in MIDI format), run:

python inference.py --ckpt_path PATH_TO_MODEL --input_file PATH_TO_INPUT_MIDI --output_file PATH_TO_OUTPUT_FILE

Citation

@article{tang2023reconstructing,
  title={Reconstructing Human Expressiveness in Piano Performances with a Transformer Network},
  author={Tang, Jingjing and Wiggins, Geraint and Fazekas, George},
  journal={arXiv preprint arXiv:2306.06040},
  year={2023}
}

Contact

Jingjing Tang: jingjing.tang@qmul.ac.uk