My daughter sent a pointer to day to this site:
They are selling text-to-speech with different voices, alas, on Mac and Windows platforms only. I would totally buy one of these for the house.
I like to test text-to-speech programs with tongue twisters. They always get it right, of course. My favorite:
I am not a pheasant plucker, I'm a pheasant plucker's son,
and I'll sit here plucking pheasant 'til the=20
pheasant plucking's done.
It occurs to me that it would be fun, and maybe even useful, to have a TTS system that has some of the usual troubles with tongue twisters. It probably wouldn't be hard: they break speech down into phonemes or half-phonemes, and render. A bit of markov chaining, or a dictionary sorted, perhaps from the pun projects of yore.
Might this be actually useful? If you have a speaking robot. which would certainly use TTS, human-like mistakes could increase acceptability. The robot could be an abject of slight ridicule, like a really bright, good natured foreigner with an accent and an incomplete understanding of the language.