[Question] Provisioning of the Vosk model

[zahlman]

Hi, I'd like to offer a direct installation model for this project. In my model, a wheel would be published on PyPI , which could be directly installed into a self-contained environment using Pipx, and then run right away. Pipx can automatically take care of putting executables (via symlinks) onto the user's path, so the nerd-dictation command would always be available. Because Pipx creates separate "virtual environment"s for the applications it installs, there would be no risk of the installed Vosk library conflicting with the system Python installation. Users would also automatically avoid the issue with system Python installations being marked as "externally managed" - as described by kohane in #103 , and also related to #120 , #135 and probably more.

I have a quick question about the current installation instructions. Currently, users are expected to download a Vosk model (from alphacephei.com) separately and unpack it as a specifically named folder in the cloned repository. My question is, is there a specific reason (e.g. something to do with licensing) why the model files are not already included in the repository? Following the installation model I have in mind, it would be much easier to include those files directly in the published wheel. It would be possible to automate the download (in Python) from the install process, but running custom code at install time may scare away some users and adds overhead (see this blog post from a Pip maintainer and Python core developer to understand in more detail).

[KIMWW]

+1