Back in the days of 1 MHz 8-bit CPU personal computers (Apple II, Atari 800, et.al.), there were software programs that could do comprehendible arbitrary text-to-speech synthesis on those PCs. What published speech synthesis algorithms might be suitable for implementation on micro-controllers with similarly limited performance and memory? (If any...)
The methods I played with on 8-bit home computers back in the 80's involved having a small set of phonetic samples and then having the text you want to speak defined in terms of those samples.
Since I didn't know about IPA at the time, I just used two sounds per letter and allocated the short sound to lower case letters and long sounds to the uppercase. This did leave quite a few gaps though, o for /ɒ/ (hot) and O for /oʊ/ (open) were easy, but /uː/ (the oo sound in moon) couldn't be represented in my system, which was a problem for me, given my name (/buːð/ rhymes with /smuːð/, try them in itinerariums phoneme synthesis web page).
These days, you could do much more clever stuff. For a start you could use lookup techniques to translate known words in orthographic text into your known phonetic symbols, and for unknown words, implementing Phonics as an algorithm should do a reasonable job.