I note that text to speech has come a LONG way. It's not perfect, but it's better than microsoft Sam.
That said, I use a LOT of procedural text generators in my games, and it would be trivial to add markup to the generator files that would silently produce the corresponding SSML in the background. The next logical leap is to use a Text to Speech engine like Amazon Poly to generate voices for in-game characters (and put a large portion of the amateur voice acting crowd out of business - sorry!), but there's two issues: one …