Those "herky-jerky" TTS systems are considered "low-end". They really do suck.
You can by "high end" TTS systems for a lot of money, and I think with a little work there could be an open source "high end" engine as well.
I think somebody experimented in using neural networking to improve TTS (to try to get it to mimic a speech sample of a human), and the result was reasonably natural sounding.
Mark
On 25/04/05, Andrew Lih andrew.lih@gmail.com wrote:
On 4/26/05, Mark Williamson node.ue@gmail.com wrote:
And I should add, why do we need a spoken Wikipedia? Wouldn't it be better to adapt a TTS engine to get articles from Wikipedia, and read them? Remember, TTS engines can be made for any language, and it's becoming increasingly easy to make them even if you're not an expert.
Mark, not sure if you're wondering about the whole idea of audible aritcles, or just the human-centered approach to it.
If you find a TTS system that is acceptably good (Festival, MBROLA, et al) then feel free to try it and post some samples at [[Wikipedia:WikiProject Spoken Wikipedia]]. But so far the TTS results I've heard have been quite unsatisfactory, and the "ear fatigue" experienced when listening to a herky jerky TTS system doesn't bode well for long articles. Wikipedia might be a great testbed for a TTS system. I'd like to see how it does on mixed language articles like [[Dim sum]].
-User:Fuzheado _______________________________________________ Wikipedia-l mailing list Wikipedia-l@Wikimedia.org http://mail.wikipedia.org/mailman/listinfo/wikipedia-l