[bug#58140] Simple offline text-to-speech incoming !

[Nicolas Graves]

Hi guix!

I've been trying to build python-nerd-dictation for guix for the past
few days, and eventually managed to get a working package.

I'm quite enthousiast about this one, I think it makes a great addition
to the distribution, but it also raises some questions.

1) The wayland version relies on ydotool. Which is not packaged yet, so
https://issues.guix.gnu.org/50847 is a prerequisite for this patch
series.

2) Vosk-API relies on models. I was thinking about the opportunity to
also package these models in guix, since it's only a copy and would make
it more guixy. BUT the big downside is that some of these models are
huge (~2G). Is there a way to not provide substitutes in some such cases
? Maybe that's a thing we can do for all packages relying on a
copy-build-system or something like that ?

3) The wayland version requires some configuration in order to work, for
ydotool particularly. For now, I keep this configuration in a "feature"
from Andrew's Tropin RDE (see
https://lists.sr.ht/~abcdw/rde-devel/patches/35600), but it might be
smarter to try and put that in a service for guix. WDYT ?

I'm going to check everything builds before sending patches, so it might
take an hour or so before patches arrive.

-- 
Best regards,
Nicolas Graves