What I don't understand is why text-to-speech is so bad (on the iPhone at least), and it's related to this editing issue. Text to speech will never be a perfect replacement for other forms of text entry, because you're limited in when you can use it. But it's so close to being good.

The main issues are so easy to solve, they're just silly UI issues around editing. You can enter text verbally, it works surprisingly well at understanding what you're saying. But... you can't delete the words you just said if it was wrong. It's literally the first thing you run into - you're transcribing, it gets one word wrong, boom, you're stuck in normal tap-editing-land again. Trying to enter messages while driving? Good luck, unless you can transcribe the message perfectly on the first go, you have to use your hands.