OCR is just a matter of time, faster than speech recognition. All it needs is to just understand language and context to self-correct. Even if that doesn't mature well enough, I think the small mistakes are okay for manual edits :)