This is a techdemo, not a tutorial, but it''s cool : http://www.research.att.com/~ttsweb/cgi-bin/ttsdemo

If you really want to cheat you could use Winsock to do your test to speech... understand?

quote:

---

Original post by Michalson
If its simply the functunality that you want (ie. its not a school project on speech synthesis) then you could just download
Microsoft Speech SDK.

---

HAHAHA, right, like my pidly little 56K connection can download that in any reasonable amount of time. lol, j/k. Thanks for the suggestion though, but I think I''ll enjoy the challenge of getting something like this to work on my own. And now that I have alot of suggesstions from everyone that replied, and probably more on the way, I think I can make a fairly decent implementation.

I''m going off to college this fall, do you think if I worked at it for most of the summer I could get something working pretty good that will at least make a good foundation to build off of?

Thanks guys.

-AJ



C:\DOS
C:\DOS\RUN
RUN\DOS\RUN

-Comic Book Store Guy''s t-shirt that I saw on the Simpsons, although it didn''t actually come from the Simpsons.

http://vdsoft.netfirms.com/home.html

The only way I know of doing this without years of research (the first text-to-speech program took three years to write, and I'm not sure how much information on hard coding such a program has been made public since then) is to use neural nets. Basically, you want to have a recording of each phoneme, then create a data set that defines the sequence of phonemes for a given sequence of letters. If the set is sufficiently good, then you'll have a working text to speech in as long as it takes to train the net. However, actually creating that data set isn't trivial, and that's the reason that you don't see more people try it.

[edited by - Mordoch Bob on April 15, 2002 5:57:59 PM]

But what about extending my idea to include anticipatory measures that anticipate how what sound you should play for a given letter or group letters based on the placement of not just the next letter, but of all letters in the word. I know I''m not making much sense right now, cause I''m basically just overcomplicating what I said earlier, but for some reason it''s making perfect sense to me. I''ll see if I can sketch out some kind of flow chart to show what I''m talking about. Take care.

-AJ

C:\DOS
C:\DOS\RUN
RUN\DOS\RUN

-Comic Book Store Guy''s t-shirt that I saw on the Simpsons, although it didn''t actually come from the Simpsons.

http://vdsoft.netfirms.com/home.html

Take a look at implementations of neural networks, particularly the backpropagation variety. They are quite able to do what you are talking about, which if I understand properly is using word groups to produce phonemes rather than just letters. Incidentally, this isn''t just a good way of doing it, but the only way of doing it, since most phonemes are generated from multiple letters.

quote:

---

Original post by Mordoch Bob
Incidentally, this isn't just a good way of doing it, but the only way of doing it, since most phonemes are generated from multiple letters.

---

Just wondering, but from what experience does this come from? I know I'm an AI newbie, but it's just that what I'm thinking makes perfect senese to me and doesn't seem too terribly complicated. I mean, if the neural nets are the only way to do it, then there's obviously a flaw in my thought process. Can you tell me where my thinking has gone awry so I don't make the same mistake in the future? Thanks.

-AJ



C:\DOS
C:\DOS\RUN
RUN\DOS\RUN

-Comic Book Store Guy's t-shirt that I saw on the Simpsons, although it didn't actually come from the Simpsons.

http://vdsoft.netfirms.com/home.html

[edited by - redneckCoder on April 17, 2002 10:12:57 AM]