Yet Another Audio Splitter

This is Yaas 1.0.0, a desktop application that splits the soundtrack of a YouTube video into separate stems (vocals, drums, bass, other, ...), for example to practice an instrument over the rest of the band.

Browse to a video in the built-in browser, click Start, and Yaas downloads its audio, separates it with a machine-learning model, and writes one WAV file per stem.

Full documentation: https://kleag.github.io/yaas/

Installation

Ready-to-run builds are on the GitHub Releases page:

Platform	Download	ffmpeg
Windows	yaas_installer.exe	Install it separately: winget install ffmpeg in PowerShell
macOS (Apple Silicon)	yaas_installer.dmg	Included
Linux (x86_64)	yaas-x86_64.AppImage	Install it with your package manager, e.g. sudo apt install ffmpeg

The macOS app isn't signed: the first time, right-click it and choose Open to get past Gatekeeper's warning. On an Intel Mac, install from PyPI instead.

On any platform with Python 3.12 to 3.14, you can also install from PyPI into a virtual environment (see the uv documentation), with ffmpeg installed separately:

uv pip install yaas

See Installation for details, including GPU acceleration.

Usage

Start Yaas from your applications menu, or with yaas in a terminal when installed from PyPI. Then:

1. navigate to a YouTube video or playlist in the integrated browser,
2. click Start and wait: separation can take a while, especially on CPU,
3. click the stems listed in the status log, or find them in the output folder, $HOME/yaas_tracks by default.

While a job runs, keep browsing and click Add to Queue to split more videos afterwards. Click Stop to stop the running job.

The ☰ menu gives access to:

• Settings...: the output folder and the separation model,
• Open Output Folder,
• GPU Acceleration...: an optional CUDA environment on Windows/Linux, or the status of Metal acceleration on Apple Silicon Macs,
• the documentation, the issue tracker, and the version information.

Separation models

Choose the model in Settings...; the choice is kept for future runs.

Model	Library	Notes
BS-Roformer (default)	audio-separator
HTDemucs 6 stems	audio-separator	Also separates guitar and piano
OpenUnmix	OpenUnmix

Models are downloaded on first use and cached for later runs.

Command-line options

These override the settings for a single run:

Option	Description
--version	Print the version and exit
-o, --out DIR	Output folder
--backend {audio_separator,openunmix}	Separation library
--model {roformer,htdemucs6s}	Model used with the audio_separator backend
--sample-rate {44100,48000}	Stems' sample rate

Please respect the copyright of the videos' authors: if they don't allow sharing, keep the extracted stems for your personal use.

Development

Yaas uses uv. From a clone of the repository:

uv venv && source .venv/bin/activate
uv pip install -e . --group dev
yaas
pytest

Releases are versioned with bumpver:

git commit
bumpver update --patch   # or --minor / --major

bumpver pushes a version tag, which makes GitHub Actions publish the package to PyPI, and build the Windows, macOS, and Linux packages and attach them to a GitHub Release. See Building & Releasing for building them locally.

Author and license

Gaël de Chalendar, aka Kleag (c) Gaël de Chalendar, 2024-2026

This program is free software, licensed under the Mozilla Public License 2.0 (MPL 2.0) license (see the LICENSE file). It includes most of the youtube-to-mp3 project, itself under the MPL license.