There is a fantastic freely-available (X11-like license) Text-2-Speech tool called Festival that you might check out. I believe the majority of the code is built in libraries, which at least in theory would make it easier to build into the Empeg. (i.e. just throw the appropriate hooks into the player software to call the Festival API calls. It's a simple matter of programming, you see...)

It's possible that whomever it was a couple messages back mentioning a T2S software that could mimic personalities was thinking of Festival. You can, if you are so compelled, build "phoneme libraries" from any voice you want to take the time to sample and take apart and Festival will then use that voice to speak to you. They still sound pretty computerized, but noticably less mechanical than Stephen Hawking's chair for example. :) (To help those crazy^H^H^H^H^Hdedicated enough to want to build their own speech libraries, there's a thing called Festvox to help you.)

(When looking up Festvox just now, I also ran across an interesting-looking project called Knight Rider.)

I agree with probably everyone else that were a feature like this available, it would have to have an "On/Off" option. I also agree that it could be extremely cool.