MetricGAN with PyTorch, currently in progress. I will finish this project in the near future 🚧 🚧
- Spectral normalization. Official uses a Keras-based implementation, but in the current repository we use the official implementation of PyTorch.
- Batch size. official only support batch size equal 1.
- Install CUDA and PyTorch
pip install -r requirements.txt
TODO
- Add generator and discriminator
- Add optimizer and data preprocess
- Add config file
- Add trainer and train script
- Add validation logic
- Add normalization of inputs
- Add clipping constant
- Do more tests
- Improve comments and README
- Add enhancement script