This full duplex spoken thing, it's already for quite a long time being used by the big players when using the whatever "conversation mode" their apps offer, right? Those modes always seemed fast enough to for sure not be going through the STT->LLM->TTS pipeline?