Machine speech

so, the situation: i'm designing a language for a computer game, one that
will be spoken productively by characters in-game (the grandiose plans are
for a very open-ended roleplaying setting, where the AIs will respond
based on your open-ended actions.  but that's another story.)

anyway, this means that it will need to be machine-produceable.  i have no
experience with this, and was wondering what kind of phonology would be
well-suited to unit-selection or diphone synthesis.  also, as i'm creating
the phonology, is one of those two methods computationally easier, and
should i therefore make a phonology suited for it?

any leads would be greatly appreciated.

thanks in advance,

kit