A customer of mine has a product based on DragonNaturallySpeaking that medical doctors use for dictation. They are doing a brisk business. It's all software.
No fancy hardware. For this use case, it's almost flawless. And way faster than typing. The only commands you give are related to the recording -- stop/start/playback-type commands. Not sure how sophisticated it gets beyond that. But the speech-to-text technology is pretty damn good now.
At this point in the SOTA, Whisper [0], can probably be a drop in replacement for the for-profit Nuance versions of Dragon, etc. It's even open source.