Differentiable Wave Digital Filters

This repository contains an exploration of implementing differentiable wave digital filters (WDFs), in an effort to integrate WDFs with neural networks, and allow for the automatic optimisation of WDF parameters.

Organization

The repository is organized as follows:

diode_dataset/  # Dataset used for training models of diode circuits
modules/        # Third-party libraries
plugin/         # Audio plugin (JUCE/C++) containing real-time WDF models
wdf_py/         # Differentiable WDF library, and scripts for training WDFs

Settings up the Python environment

The code in this repo has been tested using Python version 3.9.5. While it is possible to just run this code on your machine as-is, we reccomend using virtualenv to help manage dependencies and versions.

# create virtualenv (only need to do this once)
virtualenv --python=python3.9 env

# enter virtualenv
source env/bin/activate

# install requirements
pip install -r requirements.txt

# do your stuff...

# leave virtualenv
deactivate

Building the Audio Plugin

The audio plugin can be built using the CMake build system.

# Clone the repository
$ git clone https://github.com/jatinchowdhury18/differentiable-wdfs.git
$ cd differentiable-wdfs

# initialize and set up submodules
$ git submodule update --init --recursive

# build with CMake
$ cmake -Bbuild
$ cmake --build build --parallel 4

If you'd like to make an optimized "release" build, it is suggested to use some slightly different build commands:

$ cmake -Bbuild -DCMAKE_BUILD_TYPE=Release
$ cmake --build build --config Release --parallel 4

The resulting builds can be found in the build/plugin/DifferentiableWDFs_artefacts directory.

Citation

If you are using this code as part of an academic work, please cite the repository as follows:

@InProceedings{chowdhury:clarke:diffwdfs:2022,
    author = {Jatin Chowdhury and Christopher Johann Clarke},
    title = {Emulating Diode Circuits with Differentiable Wave Digital Filters},
    booktitle = {19th Sound and Music Computing Conference},
    year = {2022},
    pages = {2-9},
    url = {https://zenodo.org/record/6566846},
}

License

The code in this repository is licensed under BSD 3-clause license.