I expected TTS/STT to be a solved problem now with decades of work on it. I can understand LLMs having greater latency but all flagship smartphones have infere

I expected TTS/STT to be a solved problem now with decades of work on it. I can understand LLMs having greater latency but all flagship smartphones have inference accelerators these days. A good response is just an API away, which Google Assistant already does I think (it doesn't give me an instantaneous answer ever).