How much latency is acceptable? If you're off in the woods somewhere far from the cloud, sure, but it's less than 10ms to ping Google.com for me, and if the speech-to-text engine runs faster than realtime, I don't see why processing remotely is a problem. 10 ms is nothing.
Still, the transcription part is already here today. The Google Translate app has a transcribe app that does this (runs locally; does not do magic AI "pick voice out from crowd"). My father-in-law has been using it for years. When I'm in a loud environment, the app I use on iOS is called Big, which just displays large text on the screen.