/TFlite-VSTplugin-template

JUCE Template plugins to use TensorFlow lite for deep learning inference

Primary LanguageC++

TFliteTemplatePlugin

This is a template plugin for TFlite framework. It is intended to be used as a starting point for creating new plugins. If compiled as it is, the plugin will just pass the audio through. However, it will prove that a plugin that integrates TensorFlow Lite can be compiled and run on either the local machine or on the Elk Audio OS.

This is part of the elk-audio-AI-tutorial, where you'll find working examples to be compiled.

Crosscompilation has been tested as a headless plugin for the Elk Audio OS on a Raspberry PI4.
/libs includes an TensorFlow Lite (v2.11.0) binary for linux-aarch64.

Instruction for Elk Audio OS

  1. Download and install the Elk Pi4 SDK for the desired OS version .
  2. cd into libs/ and run the TensorFlow Lite compile script for the desired architecture.
  3. Open TFliteTemplatePlugin.jucer with the Projucer and save to generate the project files.
  4. cd into Builds/linux-aarch64 and run compileDebugElkPi4.sh (Making sure you downloaded the appropriate ElkPiSDK, read the script comments for more info).
  5. Copy the generated Builds/linux-aarch64/build/TFliteTemplatePlugin.vst3 to your Elk Pi4 in the ~/plugins/ folder.
  6. Copy the desired JSON config file from config_files/ to a folder of your choice on the Elk Pi4.
  7. From a terminal in the board, run: 
    sushi -r -c /path/to/config_tflitetemplate.json

This is part of the elk-audio-AI-tutorial.

2023 Domenico Stefani