[MUD-Dev2] [Offtopic] Lend your ears to science!

[Mike Rozak]

For many years now, I've been posting that (in the long run) text-to-speech will be an important technology for MMORPGs.

If you want to listen to the latest bleeding-edge text-to-speech research, AND at the same time help improve text-to-speech, then please run through the listening tests at http://homepages.inf.ed.ac.uk/mfraser2/blizzard2007/register-R.html .

A quick explanation of what's going on in the test:

Modern text-to-speech voices are made by taking several thousand recordings of someone speaking, analyzing them, and then producing a voice file. In the case of the blizzard test, 6000 recordings were sent out to 16(?) different companies a few months ago. The voices were then used to synthesize the test sentences that you hear.

In each section you'll listen to one sample from each company's voice, as well as one sample directly from the original speaker. You then have to give the sentence a score about how realistic it is, or type in what you heard, depending on the section.

These scores are then tabulated, and text-to-speech engines are ranked. (Mine will be near the bottom this year, but I'll get it better for next year.) Participants then write papers describing what they did, and use each others papers to improve their algorithms for the following year.

You might find the test interesting because a few of the 16 companies have produced voices that are really good... although they still sound like a bored telephone operator. :-(

PS - Forward this around, since the more participants, the more accurate the tests.