This is the end-user install guide for running aTrain on Linux via the command line. For the recommended install, use the Flathub package. The manual setup below is for systems where Flatpak is not an option, or for users who prefer a virtual-environment install.
Developing aTrain on Linux? See CONTRIBUTING.md, which covers the uv-based workflow, the native GUI build dependencies, and the Docker dev container.
- Flathub (recommended): https://flathub.org/apps/io.github.juergenfleiss.aTrain
- A
.debfile for beta 1.2.1 is available on the university download page.
- No Snap package is currently provided. Beyond Flathub, installation and start are via the command line, but with the aTrain user interface.
- CPU transcription is the reliably supported path on Linux.
- Additional packages are installed via
aptfrom the Universe repository. - Tested on Ubuntu 24.04 LTS.
Update package lists:
sudo apt updateInstall dependencies:
sudo apt install ffmpeg python3 python3-pip python3-venv git build-essential \
libgl1-mesa-dev libcairo2 libcairo2-dev libgirepository1.0-dev \
libwebkit2gtk-4.1-dev -yCreate a virtual environment in your home folder (or any other you prefer) and activate it:
cd ~
python3 -m venv atrain_venv
source atrain_venv/bin/activateInstall aTrain (with the GUI extras) from the GitHub repository:
pip install "aTrain[gui] @ git+https://github.com/aTrainTranscription/aTrain.git@develop"On Linux the PyPI PyTorch wheel already bundles CUDA, so no extra package index is required.
Download the models for transcription and speaker detection. This only has to be done once:
aTrain initStart aTrain. This command opens the user interface:
aTrain startIf aTrain is already installed, start it with:
cd ~
source atrain_venv/bin/activate
aTrain startDepending on your Ubuntu version you might need a newer Python than the system default. Install it from a third-party repository, then run the commands above with that interpreter:
sudo add-apt-repository ppa:deadsnakes/ppa
sudo apt install python3.12