PhraseGEN: Phrase-Level Symbolic Music Generation

This is the code repository for our paper Phrase-Level Symbolic Music Generation.

Dataset

You may download the digital scores in kern format and prepare the Sonata Dataset used in the paper, or prepare your own dataset.

Prepare the Sonata Dataset

1. Data collection: download the musical scores in the Humdrum **kern data format. See sonata-dataset/README.md for more details on the classical piano sonata collected in this dataset.
2. Preprocess: See ./src/preprocess/README.md for more details.

Training

```
python3 train.py
```

Inference

1. Prepare test set with the test samples in sonata-dataset/example_test_info.json.

   python3 prepare_testset.py
   

2. Inference

   python3 inference.py
   

Citation

If you find this repo useful, please cite our paper.

@article{Feng2024Phrase,
	author = {Feng, Yijing and Sahin, Egemen and Casey, Michael A.},
	journal = {AIMC 2024 (09/09 - 11/09 )},
	year = {2024},
	month = {aug 29},
	note = {https://aimc2024.pubpub.org/pub/2e7q3cpr},
	publisher = {},
	title = {Phrase-{Level} {Symbolic} {Music} {Generation}},
}