Sonic: A Low-Latency Voice Model for Lifelike Speech