“To achieve the fidelity of a human-to-human conversation, voice AI systems need to reduce their end-to-end latency to a threshold of approximately 200 milliseconds.”