Which text to speech model sounds best? Cartesia's Sonic 3.6, first on the Artificial Analysis Speech Arena. Then come the ...