The sad truth is, voice synthesizers suck.
When I was a kid, we had this bondwell CP/M computer with a cool voice synth. If you've heard U96's das boot, it's exactly the same synth. I remember seeing some magazine ads for tiny synth chips you could put into any home appliance, and that in the future all devices would speak to us. We even bought a voxbox synth for our speccy back then. The future seemed bright.
Then, nothing.
Couple years ago I did some research on the current state of voice synths, and it's depressing. Text to speech has improved, but the voice synthesis itself? Basically no advances since then, apart from the use of bigger and bigger recorded voice sample libraries. Toss in a complicated word that's not in the dictionary, and the illusion breaks apart.
One hobby project of mine would be to design a voice synth that doesn't just concentrate on text to speech, but would let the "speech designer" to compose the tone of the speech in a more flexible way. This would be perfect for games, except that using actual voice actors would probably result in a better quality, cheaper. Of course you could use it in a mobile game where huge voice samples wouldn't fit in memory, but the budgets there are so ridiculously low anyway that it wouldn't be commercially feasible.
Still fun to think about.