Yahtzee Croshaw said:
How exactly would you say the name "Qqq%kblk?"
I would pronounce it rq-koh-k'X-keh-bu-ilk where rq is the Inuktitut style deep throated 'k' sound and 'X is a click sound from african languages like X'hosa.
Really, though, that's were the real problem with voice synthesizers come in. Different languages use very different tonalities; where in the mouth sounds are primarily formed, how open your throat is, tautness of your glottis, use of the tongue and so forth.
I have an unusual surname for where I live, so people usually give up when they can't pronounce it even after I spell it out and repeat it a few times. Good look to any speech synthesizer.
Faced with unknown data even the most advanced softwares today will stumble. Most of them still resemble expert systems in their efficacy. They are more efficient and produce a more natural sounding voice but they can't piece together pronunciation from context since they don't actually understand the content. That is, they don't have an internal, self-defined representation of that context derived only from the input and their own internal representations.
In fact, no one even knows how to do that. Every human just does it by instinct! As I have plenty of examples of though, it's far from perfect. I really don't mind when people mispronounce or misspell my name, it doesn't matter. I'm not so sure I'm willing to accept that from my computer.