torchsynth 0.0.3

A modular synthesizer in pytorch, GPU-optional and differentiable

torchsynth

The fastest synth in the universe.

Introduction

torchsynth is based upon traditional modular synthesis written in pytorch. It is GPU-optional and differentiable.

Most synthesizers are fast in terms of latency. torchsynth is fast in terms of throughput.

You will need to install the particular version of torchcsprng for your CUDA device. Please follow their simple installation instructions. But if you use the CPU version of torchcsprng, it probably won't affect performance much.

Development Installation

git clone https://github.com/turian/torchsynth
cd torchsynth
pip3 install -e ".[dev]"

Make sure you have pre-commit hooks installed:

pre-commit install

This helps us avoid checking dirty jupyter notebook cells into the repo.

Note that torchsynth requires PyTorch version 1.7 or greater.

Examples

Unfortunately, Python 3.9 (e.g. OSX Big Sur) won't work, because librosa repends upon numba which isn't packaged for 3.9 yet. In which case you'll have to create a Python 3.7 conda environment. (You might also need to downgrade LLVM to 10 or 9.):

conda install -c conda-forge ipython librosa matplotlib numpy matplotlib scipy jupytext
conda install -c anaconda ipykernel
python -m ipykernel install --user --name=envname

and change the kernel to envname.

Tests

Unit testing is performed using pytest.

pytest and other project development dependencies can be installed as follows:

pip3 install -e ".[test]"

To run tests, run pytest from the project root:

pytest

To run tests with a coverage report:

pytest --cov=./torchsynth