This is a simple addon for Docspell to add support for audio files. It uses external tools like ffmpeg and stt to extract text from an audio file. It uses wkhtmltopdf to create a pdf file.

The result is a preview + pdf and indexed text in the Docspell dms.

Note: The accuracy of text extraction depends on the used models. For simplicty this addon uses pre-trained models. For better results, you need to train your own models.

This addon supports these runners: nix-flake and trivial. An internet connection is required to download the model files.

It is recommended to install nix on the machine running joex. This allows to use the nix-flake runner which can build the addon with all dependencies automatically.

Otherwise, for the trivial runner, you need to install these tools manually: curl, ffmpeg, stt (v0.9.x) and wkhtmltopdf. The latter should already be available since it is a requirement for joex itself.

Only x86_64 archtiecture is supported, because nixpkgs doesn't provide stt for other architectures.

Currently there is nothing to configure. Just install the addon and add it to a run configuration.

Install direnv and nix and allow the source root via direnv allow. This applies the devShell settings from flake.nix. Then build the addon:

nix build

Now you can run it:

./result/bin/audio-files-addon

It will run on the test files provided in test/ and put results in test/tmp.

For quicker turnaround you can also run the source file itself. This works, because devShell puts all required binaries in path.