As many of you know, I'm a fan of using text-to-speech for CRPG/MMORPG/IF games, particularly NPC speech... basically because TTS is a LOT cheaper than hiring voice talent, AND TTS can speak anything generated at run-time.
Every year, text-to-speech researchers around the world enter a friendly competition called the Blizzard Challenge. To test the quality of their voices, the researchers provide a public listening test (which is where you come in).
Participating researchers and companies are given approximately 10,000 sentence-length recordings of a real person, and told to generate a voice from that. Normally, the original recordings are spoken in professional newscaster-style speech (proper and boring). This produces a proper and boring voice... aka: One that sounds like a "bored telephone operator".
This year's challenge is particularly challenging because the voice is a British speaker who speaks with a lot of personality (much more than your typical BBC presenter). Such a voice provides additional challenges to the researchers' algorithms, BUT if successful, it produces a voice that doesn't sound permanently bored... which is good for games, where dead-pan speech ruins the experience.
Please participate in this year's listening test at:
http://groups.inf.ed.ac.uk/blizzard/blizzard2008/english/register-ER.html
If you know Mandarin, you can also try this test with a Mandarin speaker:
http://groups.inf.ed.ac.uk/blizzard/blizzard2008/mandarin/register-MR.html
The overall test is divided into 5 sub-tests. Each sub-test has you listen to around 21 sentences. Each sentence is from a different text-to-speech engine, with one sentence being the original speaker. You will be asked to rate the quality of each sentence or type in what you think was spoken.
Some of the TTS engines are fairly poor (mine included, although it's much better than last year). Some are quite good (IMHO that is; you be the judge). While you're listening to the tests, imagine using the better TTS engines in your game.