I've always thought that was a really bad idea for general purpose computing. Can you picture an office with everyone yakking at their computers? And a decent typist can type faster than they talk for many things. While you might be able to dictate a standard document, trying to do complex things like coding just doesn't work. Even if, someday, we had AIs which could understand speech 100% of the time - dictating code to another person sucks, doing it all day to a computer would just suck more. And how are you going to work on a computer while being on the phone - think call centers, tech support, etc. Or what if you like to listen to music? Or if there are other people in the area talking?
It just doesn't strike me as a good idea, which is why it remains The Future!™. Now, for specific tasks like car navigation systems, phone dialing programs, etc, it makes sense. Situations where speech is the most logical interface - like a phone or a car, where you don't want the person looking at a screen and poking at it while driving. And it works well enough there because it doesn't have to accept general speech, but a limited set - much like Graffiti on PalmOS worked better than most handwriting recognition systems, because it forced the user to write in a specific format - not anyway they pleased.
Why am I talking about this? Because this video had me in tears from laughing so hard.
Video ganked from jwz and mackys