Skip to content

Latest commit

 

History

History
44 lines (38 loc) · 2.68 KB

README.md

File metadata and controls

44 lines (38 loc) · 2.68 KB
Raw

DIST-NN - A parametric distortion neural effect

Introduction

In this project for the "Selected Topics in Music and Acoustic Engineering" course, we presents DIST-NN, a parametric neural effect for distortion. The Plug In is implemented in C++ using the JUCE framework for reliability and speed. For what concerns the management of the neural networks, we decided to go with RTNeural, a lightweight neural network inferencing engine.

Training

For the training procedure, we used Google Colab for generating the models and Google Drive for storing the .json files. The training pipeline is the following:

  • clone this repository
  • upload the "TRAIN" folder on Google Drive
  • in the "parametric_data.txt" file you can find the data folder for the training procedure
  • open the trainParametric.ipynb in Colab
  • insert the paths for your "TRAIN" folder and your "parametric_data" folder
  • train with the proper terminal command

GUI

For what concerns the Graphical User Interface, it is composed by:

  • an On/Off button that enables or disables the effect
  • a knob that controls the amount of effect to apply (in our case, the distortion)

DIST-NN_GUI

Requirements and Usage

Requirements

  • clone this repository
  • download CMake here: https://cmake.org/download/
  • download RTNeural here: https://github.com/jatinchowdhury18/RTNeural
  • download JUCE here: https://juce.com/download/
  • put the RTNeural folder inside the repository folder
  • open the CmakeLists.txt file, and insert your JUCE path at line 7
  • copy and paste the DistNNutils folder to your Deskstop (this is just for showing the background image)
  • from terminal, go to the plug in folder and run: cmake --build build --config Release
  • from the build folder created in the previous point, go to the Release folder and find the .vst3 file

Usage

The default model is the one using 16 hidden layers. In case you want to try the other models (8, 24 or 32 layers) you have to edit the code:

  • in the src folder, open the PlugInProcessor.cpp, at line 25 the 16 hidden layers model is loaded. If you want to change model, substitute "modelParametricDIST16_json" with "modelParametricDIST8_json", "modelParametricDIST24_json" or "modelParametricDIST32_json"
  • in the src folder, open the PlugInProcessor.cpp, replace 16 with either 8 or 24 at line 191
  • in the src folder, open the PlugInProcessor.h, replace 16 with either 8 or 24 at lines 57 and 63

Links