mbrola is also mentioned as source for voices in the article, @Rosika cited.
I’m not able to provide any solution to the original problem, but I’d like to take the opportunity to mention the Mozilla Open Voices project:
Every person can contribute with her own voice in her own language. I believe, this project is important. Whilst there exist pretty good commercial applications for voice recognition and voice synthesis, only an open project with a huge voice database can provide the base for academic research in the field and open source applications.
If you are a woman, I would especially encourage you to participate: In all languages, there are far fewer female than male voices in the database, but who would want to hear her computer talking in a male voice?