Music-Style-Transfer
This is music style transfer with tensorflow 2.0. One of the architectures is based on [1] and [2], but we implement it from scratch.
Original:
ResNet (Jazz):
LSTM V2 (Jazz):
Our Environment
- Operating System:
- Archlinux/Windows10
- CUDA:
- CUDA V11.5.50
- Nvidia driver:
- 495.44
- Python:
- python 3.7.3
- Tensorflow:
- tensorflow-2.6.0
Downloading the Preprocessd Dataset
Downloading the Pretrain Model for Classifier
Train MusicStyle Transform with CNN
cd ./MusicStyleResNet
Train MusicStyle Transform with CNN and evaluation
cd ./MusicStyleResNet
python main.py --load_classifier ../Classifier/checkpoints/Classifier/{timestamp}/{checkpoint}.hdf5
Continue Training
cd ./MusicStyleResNet
python main.py --load_checkpoint ./checkpoints/ResNet/{timestamp}/{000-0.250-1}
Test
cd ./MusicStyleResNet
python main.py --load_classifier ../Classifier/checkpoints/Classifier/{timestamp}/{checkpoint}.hdf5 --load_checkpoint ./checkpoints/ResNet/{timestamp}/{000-0.250-1} --phase test
Generate Sample Output
cd ./MusicStyle{model}
python main.py --load_checkpoint ./checkpoints/ResNet/{timestamp}/{000-0.250-1} --phase sample --sample-midi {midi filepath}
Train Genre Classifier
cd ./Classifier
python main.py
Test Genre Classifier
cd ./MusicStyleResNet
python main.py --phase test --load_classifier ./checkpoints/Classifier/{timestamp}/{checkpoint}.hdf5