New chat app idea: You "text" a locally running TTS model that uses your phone's modem to call the recipient and talks to the recpient, and replies (from another TTS model on the other side) are run through STT model and appear as text replies.
But pushing this absurdity a bit further, what exactly makes it a voice call instead of a text chat? Either way, there are bits moving over wires.
Let's make a new "voice codec" that can trivially encode and decode to text (like Speak'n'spell phonemes) so that the bits on the wire technically represent some "audio data", but with much less overhead than the usual TTS.