You can install the app directly from the original developers, although it’s a confusing mess to do so.
But it allows access to non-piper voices which I find to sound better.
For example, conqui AI voice 80 is my favorite.
Ohhh boy SherpaTTS got an easy fdroid install! That’s a game changer, nothing to be scoffed at compared to Google’s… And fully in device/offline. I’m still trying to figure which model sounds best, when I was beta-testing it last year I don’t think they had this many models. Also the model installation is just slightly less janky than before, but still… Wanna change model? Delete the storage or reinstall the app!
Yeah, it could definitely use some work. So far, my favorite is English HFC female.
Sigh… I don’t know which one i installed now because it doesn’t even tell you!
As a talk back user, this is really handy. However, I will say that these speech models do not work very well at high speech rates, which is what I need for a screen reader user.
It is definitely much better quality, though. That’s for sure.
Also, for some reason it seems to skip letters sometimes. As an example when I turn on TalkBack foss or turn off TalkBack foss, it says “alkback foss turned on” or when i first opened thunder (lemmy client) it said “under”. Or at least that’s what it sounded like.
I will have to play with different models, but so far my favorite is English HFC female.
I am not sure whether I will use this as my day to day text a speech engine with my screen reader as of right now but I will definitely be keeping a close eye on this one because I really do like what I hear so far.
Edit: It could be that the model is trying to start talking before the speaker is ready, because it seems like it only has a big effect when you’re moving to a new element in TalkBack, which means that the system has to turn on the speaker first.
