The missing UX for voice transcription is a way to easily edit while you're talking. Even just to add punctuation. I guess voice transcription and LLM prompting are a perfect match because the grammar/typos matter a lot less.