WaveNetVA

Feedforward WaveNet for black-box virtual analog modeling. This code is related to our paper submitted to SMC 2019: https://link-to-pre.print

A real-time implementation of the model built using JUCE. The code can be built as a standalone audio application or as an VST3, AU or AAX etc. plugin.

Audio samples are available on the demo page.

Getting Started

Installing

• Clone the repo.
• Download and install JUCE
  • The software has been tested with Juce version 5.4.2
• Download Eigen
  • Extract Eigen somewhere, e.g. your home directory.
  • The software has been tested with Eigen version 3.3.7
• Open WaveNetVA.jucer file with Projucer
  • Add the Eigen folder to "Header Search Paths" in Exporters -> Debug/Release
  • Open and build project in XCode or Visual Studio.
  • Remember to switch the build target to Release (using all optimizations enables realtime use)

Loading trained models

The trained models are stored as json files. The pre-trained models of the Ibanez Tube Screamer, Boss DS-1 and Electro-Harmonix Big Muff Pi described in the paper are included in the Models directory.

Training new models

Although we're currently unable to release our original training scripts, the amazing people on the Internet have reproduced the training in PyTorch and even created a compatible model exporter.

Check the following repository for more information https://github.com/keyth72/pedalnet

License

This project is licensed under the Apache License, Version 2.0 - see the LICENSE file for details.