Could you share a bit more about how you’re planning to use session.generate_reply() in your STT->LLM->TTS pipeline? For instance, are you working with real-time audio streams, or is it more of a batch processing situation? Knowing this will really help me understand the best way to integrate it for you.