Play as You Like: Timbre-Enhanced Multi-Modal Music Style Transfer

Paper

Chien-Yu Lu*, Min-Xin Xue*, Chia-Che Chang, Che-Rung Lee, Li Su, "Play as You Like: Timbre-Enhanced Multi-Modal Music Style Transfer", AAAI 2019

This is authors' pytorch implementation of the paper.

How to Use

Dependency

python >= 3.6
pytorch >= 0.4.1
librosa >= 0.6.0
pyyaml
tensorboard
tensorboardX

Preprocessing and Post processing

Code and tutorial.

Training

Prepare your own dataset.
Setup the yaml file, see configs/example.yaml for more details.
Start training.

python train.py --config configs/example.yaml

Testing

You can run test.py after finishing training process. The following line will do an a to b style translation.

python test.py --config configs/example.yaml --input dataset/pia2vio_example/ --checkpoint outputs/example/checkpoints/gen.pt --a2b 1

Results

The left two columns are the input (original) and output (transferred) features of a piano to guitar transfer while the right two columns are features of a guitar to piano transfer. From top to bottom the features are: mel-spectrogram, MFCC, spectral difference, and spectral envelope.

Audio samples for a bilateral transfer of piano to guitar and guitar to piano.

Here's the link to all audio samples.

Citation

@article{Lu_Xue_Chang_Lee_Su_2019,
  title={Play as You Like: Timbre-Enhanced Multi-Modal Music Style Transfer},
  journal={Proceedings of the AAAI Conference on Artificial Intelligence},
  author={Lu, Chien-Yu and Xue, Min-Xin and Chang, Chia-Che and Lee, Che-Rung and Su, Li},
  year={2019},
  month={Jul.}
}

Reference

MUNIT

SPSI_Python

chang810249/Play-As-You-Like-Timbre-Enhanced-Multi-modal-Music-Style-Transfer