Yeeting thoughts directly into models via voice is basically bootleg Neuralink. Give Superwhisper a try.
Harping on this point, but basically typing something usually requires knowing what you're going to say ahead of time, which isn't true for pacing around and brain-dumping through speech.