“Russ d'Sa predicts that voice AI could achieve 200-millisecond end-to-end latency and handle complex human-like repartee within 18 to 24 months.”