Quite possible, but I did some checking and even highly ranked software has ranges of accuracy that I would find poor if I was typing. Ranging from 70 - 96%. As I've discussed concerning the typical claim that 97% typing accuracy in the 60-70+ wpm range is still one error every 33 characters which seems like a lot to me. 96% would be an one word wrong out of 25 words - and that's the best. I also see things about some new version being, say, 14% more accurate than the previous one. Don't know how they calculate that. Suppose you are at 90%. Obviously it isn't based on 14% higher than 90 since that would take you over 100. So maybe it is based on the 10% error rate and they cut that by 14%. That would raise the overall accuracy to about 91.4%. Over time, yeah, that adds up, but would take a number of such improvements to be noticeable.
I do wonder if, with a reasonable amount of practice, I could speak and not have trouble with the "uh" stuff. I'm sure some can. Not so sure most will.
But the thinking process is different too. When typing, I find it much easier to think about what I want to say as I type. When talking - not conversations, but dictating - I find it really uncomfortable. Again, practice may, or may not, solve that. I'm sometimes tempted to buy one of the top packages and see, but not at those prices I've seen. And the upgrade prices tend to be darn high too! I can buy a good number of keyboards instead
As you say, all this software is improving. Dragon made a huge jump in this category a long time ago. Since then, to me, it seems to be improving far more slowly.
I'm not thinking of this as an either/or situation. In fact, I'm quite confident it won't be and that's a big reason why I don't see these programs replacing keyboards. If you can't or shouldn't speak out loud in some places, you need other options and that pretty much leaves a keyboard or writing by hand. But typing is far faster and much more legible for most people. Also, with a computing device, more practical to edit!
Your cyborg approach would be common - but partly because of the necessity to often have to search for and correct errors.
Hard to say how overall speed would be affected. Certainly one can speak faster than they type, but then you have to add in the time to make those corrections. Which is longer? No idea. Probably either could be depending on the individual.
I suspect a poor typist may do well with dictation because when typing they are so slow anyway, thus the time to correct problems isn't as big an issue in comparison.
Anyway, I like the dictation improvements and I am not arguing against it. I'm just saying that it is highly unlikely to replace the need for many many people to still need a keyboard. And the TB is a great keyboard.
While you are also correct that many will not bother with a TB when they are carrying a laptop, there is more to it. If the TB catches on (it deserves to, but it has to get people to be willing to change from legacy keyboards), I see a couple things in its favor.
One is that, like myself and many (I think most) testers, we like it so much we really don't want to go back to our old keyboards. Thus more likely to want to take it with us and use with a laptop.
Another is that some people will have some difficulty switching back and forth. This added difficulty in going back will also motivate them to keep their TB with them. Note though that some testers have found that, over time, they have no trouble switching.
And the big one - if the TB concept takes off, I can see a lot of licensing of the basic concept so we may eventually see the TB built into a laptop. Think of the spaces savings for batteries, etc! Or the tiny laptops that would no longer have very limited keyboard designs. Maybe some will even develop a laptop where this keyboard is removable so can be used with other devices. But when attached directly in the laptop, it would not use bluetooth, but would be like a wired keyboard so better for gaming.