A neural drums demixing plug-in
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.
- Download/clone the project repo
- In the
CMakeLists.txtfile, 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!
- Load your drum tracks:
- Drag and drop your external files;
- Click the load button.
- Click the
separatebutton and extract five drum kit stems:- Kick
- Snare
- Toms
- Hi-hat
- Cymbals
- Use the
play/pausebuttons 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
wavformat in two different ways:- Click the download button;
- Drag and drop the audio clips directly from the plug-in to anywhere you want!
Under the hood, LARS runs LarsNet, whose weights are distributed under a CC BY-NC 4.0 license.