Yet another WaveNet implementation in PyTorch.
The purpose of this implementation is Well-structured, reusable and easily understandable.
-
System
- Linux or macOS
- CPU or (NVIDIA GPU + CUDA CuDNN)
- It can run on Single CPU/GPU or Multi GPUs.
- Python 3
-
Libraries
- PyTorch >= 0.3.0
- librosa >= 0.5.1
python train.py \
--data_dir=./test/data \
--output_dir=./outputsUse python train.py --help to see more options.
It's just for testing. You need to modify for real world.
python generate.py \
--model=./outputs/model \
--seed=./test/data/helloworld.wav \
--out=./output/helloworld.wavUse python generate.py --help to see more options.
networks.py and model.py is main implementations.
- wavenet
config.py: Training optionsnetworks.py: The neural network architecture of WaveNetmodel.py: Calculate loss and optimizing- utils
data.py: Utilities for loading data
- test
- Some tests for check if it's correct model like casual, dilated..
train.py: A script for WaveNet traininggenerate.py: A script for generating with pre-trained model
- Add some nice samples
- Global conditions
- Local conditions
- Faster generating
- Parallel WaveNet
- General Generator