cxxpiper

[Derek Davies]

Just an update on my hacking with piper and C/C++ based SD output
module.

It speaks and cancels!  

I cribbed from the piper distro's src/cpp dir and from SD's pico.c .
Lots more to go but that is a milestone and makes it minimally usable.

I named it sd_cxxpiper, hopefully to distinguish it from the excellent
python leveraging module that people are already using (and happy with).

That said, I haven't actually used it much yet so I don't know if the
expected efficiency of C/C++ OM is visseral or not.  Happy to make my
hacks available if someone wants to test, but better to wait a week or
two.  I will post when it is cleaned up and all basic features are
there.  I have no config file parsing for instance, it's all hardcoded
right now.

Casual testing with spd-say does not incur model load time, as
expected, since that's done at module load time.

I am still interested in doing this for Coqui and I am also thinking
of hooking up google cloud Voice and windows TTS.  The lattter two I
have most of from an earlier project,.

I had to learn a lot about SD architecture, since I came into this
cold.  I kept notes on what I've learned and how my dev env works for
various things.  There is a lot that comes into play and I could not
find any "hacking overview" anywhere.  If there's nothing out there
I'm happy to start something.  No doubt there are lots of tricks
people know for debugging and testing, especially wrt systemd.  It
would be nice to have an intro to the generic output module's approch
vs a completely C based one such as pico and the range of options in
between, like festival's approch.

That's where I'm at for now.  I am so happy there are others working
on this stuff and a list to post to -- THANKS!!!

Derek

PS: Also have IBM Watson and Mimic3 stuff from a previous project that
are not so interesting to me, but would be straightforward to make an OM from.



=====


Thu Sep 12 14:27:42 2024