Finally found the time to read into this and as far as I understand it uses an external piece of HW for the speaking part and only prepares a given text into codes that then generates the speech on the mikrocontroller.

Nice, but.....
Can I just send plain text or would I need to include all those fancy phonetic symbols ?
If I don't need the phonetic stuff, how are different languages handled ? (narrator.device could only do bad english and completly failed for anything else)

Looking at that GitHub entry it seems that your program is just a rather simple/dump frontend for the mikrocontroller, if thats not the case it might be possible to create a version that just outputs to AHI (audio) with the help of some local speech-linklib (I'm sure there is something like that to be found somewhere).