I'm not sure about others but I can't coherently voice my thoughts through speech alone as I'd want to think and revise the message, so I generally don't use voice transcription with AI.

I haven't found that to be an issue. Just say what revisions you want. Once you're done, paste it into an LLM to clean it up into a usable prompt.

That sounds more annoying than just typing it myself. I guess if you're on a walk you can just talk into your earbuds and the coding agent will parse everything automatically though.