The AT&T voices do sound good. But i am afraid the text-to-speech is not useable after all for gaming (maybe in a few years time). You just have to few realy diferent voices.

But i found the solution: Pamela which i found at: http://www-personal.monash.edu.au/~m...ela/index.html
That is the solution for my problem. It allows me to:
- load recorded speech
- type the text that is spoken
- choose the language and the poneme's are transcribed over time with the recorded speech
- export a list with times and phoneme's
Now in 3DAS i can play the recorded speech and use the exported list to animate the (head/mouth) mesh.