Skip to content

Latest commit

 

History

History
60 lines (46 loc) · 2.58 KB

README.md

File metadata and controls

60 lines (46 loc) · 2.58 KB
Raw

LARS

A neural drums demixing plug-in

Description

Introducing LARS, the first open-source AI-powered plug-in for drum source separation.

Developed in Python and running in C++, LARS is a versatile tool for music producers and sound engineers alike. Thanks to its user-friendly interface, LARS makes it easy to accomplish complex tasks such as drum replacement, drum loop decomposition, audio restoration, remixing, and remastering—all within your DAW of choice.

LARS can separate a stereo drum track into five audio stems: kick drum, snare, toms, hi-hat, and cymbals.

LARS is available both as a VST3 and Audio Unit (AU) plug-in.

⚠️ LARS is currently in beta. Please help us improve it by reporting any bug. Note also that the source separation algorithm may change in later versions as we develop better drums demixing models.

Requirements

How to run LARS

  • Download/clone the project repo
  • In the CMakeLists.txt file, modify the following lines by typing in the path to your LibTorch and JUCE folders:
    • set(CMAKE_PREFIX_PATH "{your-libtorch-folder}") [Line 25]
    • add_subdirectory({your-JUCE-folder} ./JUCE) [Line 31]
  • On your terminal, go to the project folder and run:
cmake -B build .
  • This will generate a project file in the build folder, open it.
  • Compile the plug-in with your favorite IDE (we tested it only on VSCommunity and XCode).
  • Run it!

Functionalities

  • Load your drum tracks:
    • Drag and drop your external files;
    • Click the load button.
  • Click the separate button and extract five drum kit stems:
    • Kick
    • Snare
    • Toms
    • Hi-hat
    • Cymbals
  • Use the play/pause buttons or double click on the waveforms to listen to the portion of the audio files you are more interested in.
  • You can download the separated stems in a wav format in two different ways:
    • Click the download button;
    • Drag and drop the audio clips directly from the plug-in to anywhere you want!

License

Under the hood, LARS runs LarsNet, whose weights are distributed under a CC BY-NC 4.0 license.

Links