pytorch-wavenet

This is an implementation of the WaveNet architecture, as described in the original paper. Updated to work on PyTorch 1.6.

Features

Automatic creation of a dataset (training and validation/test set) from all sound files (.wav, .aiff, .mp3) in a directory
Efficient multithreaded data loading
Logging to TensorBoard (Training loss, validation loss, validation accuracy, parameter and gradient histograms, generated samples)
Fast generation, as introduced here

For an introduction on how to use this model, take a look at the WaveNet demo notebook. You can find audio clips generated by a simple trained model in the generated samples directory

Name		Name	Last commit message	Last commit date
Latest commit History 104 Commits
.idea		.idea
.ipynb_checkpoints		.ipynb_checkpoints
__pycache__		__pycache__
generated_samples		generated_samples
notebooks		notebooks
snapshots		snapshots
tests		tests
train_samples		train_samples
.DS_Store		.DS_Store
.gitignore		.gitignore
Generated_Samples.ipynb		Generated_Samples.ipynb
LICENSE		LICENSE
README.md		README.md
WaveNet_demo.ipynb		WaveNet_demo.ipynb
audio_data.py		audio_data.py
demo.ipynb		demo.ipynb
generate_script.py		generate_script.py
model_logging.py		model_logging.py
optimizers.py		optimizers.py
profiling.ipynb		profiling.ipynb
test_script.py		test_script.py
train_script.py		train_script.py
visualize.py		visualize.py
wavenet_model.py		wavenet_model.py
wavenet_modules.py		wavenet_modules.py
wavenet_training.py		wavenet_training.py