Audyn

Audyn is PyTorch toolkit for audio synthesis.

Build Status

Python/PyTorch	Ubuntu	MacOS	Windows
3.9/2.0
3.9/2.1
3.9/2.2
3.9/2.3
3.10/2.0
3.10/2.1
3.10/2.2
3.10/2.3
3.11/2.0
3.11/2.1
3.11/2.2
3.11/2.3
3.12/2.2
3.12/2.3

Installation

You can install by pip.

pip install git+https://github.com/tky823/Audyn.git

or clone this repository.

git clone https://github.com/tky823/Audyn.git
cd Audyn
pip install -e .

If you need to run recipes, add [recipes] as follows:

# In Audyn/
pip install -e ".[recipes]"

If you use MacOS, you may need to set MACOSX_DEPLOYMENT_TARGET during installation to build C++ related modules.

C++ extension

We use C++ extension to search monotonic alignment in some models (e.g. GlowTTS). To take full advantage of computational efficiency, set appropriate value of OMP_NUM_THREADS and CXX during installation:

# In Audyn/
export CXX=<PATH/TO/CPP/COMPILER>  # e.g. /usr/bin/c++
export OMP_NUM_THREADS=<SUITABLE/VALUE/FOR/ENVIRONMENT>
pip install -e "."

Development

git clone https://github.com/tky823/Audyn.git
cd Audyn
pip install -e ".[recipes,dev,tests]"

Build Documentation Locally (optional)

To build the documentation locally, you have to include docs when installing Audyn.

pip install -e ".[docs]"

When you build the documentation, run the following command.

cd docs/
make html

Or, you can build the documentation automatically using sphinx-autobuild.

# in Audyn/
sphinx-autobuild docs docs/_build/html

Test

pytest tests/package

To include slow tests

pytest tests/package --runslow

License

Apache License, Version 2.0 EXCEPT FOR WEIGHTS OF PRETRAINED MODELS
Weights for some of the pre-trained models are extracted from the official implementations. Their licenses follow the official implementations.