Write up about the architecture is here https://openai.com/index/delivering-low-latency-voice-ai-at-...