it could be done with codec2, but duty cycle of 1% would limit it to push to talk brief sentences every minute or so, it might be better to stick to text. Perhaps handy when someone had a mountain accident and can only push and talk. Use a speech to text decoder and limit expressions to 140 characters, should be enough for every piece of conversation including this comment (with vowels stripped out).
o wait, that's to little - make it 300 characters.
It would be interesting to see if you could incorporate a radio module like a DRA818U/V [1] and provide short-distance voice over MURS (VHF), FRS/GMRS (UHF), or similar bands.