Tiido's speech synthesizer 3

TmEE co.(TM) · Post by **TmEE co.(TM)** » Tue Feb 24, 2009 6:57 pm

http://www.fileden.com/files/2008/4/21/ ... PEECH3.BIN

(Lot) more work is needed around the samples, but things are shaping up

enjoy

Jorge Nuno · Post by **Jorge Nuno** » Tue Feb 24, 2009 7:37 pm

I couldn't understand a damn thing!

Could be my fault, too...

TmEE co.(TM) · Post by **TmEE co.(TM)** » Tue Feb 24, 2009 7:42 pm

You're not the only one

"1 2 3 testing testing this is demonstration of tiido's speech synthesizer bla bla bla...."

Jorge Nuno · Post by **Jorge Nuno** » Tue Feb 24, 2009 8:13 pm

How is this done? Every letter is sampled in the bin then they're glued together, looking at a word and loading the sample that corresponds to the next char?

What if the letters have different sounds in different situations?

TmEE co.(TM) · Post by **TmEE co.(TM)** » Tue Feb 24, 2009 8:19 pm

It uses 31 samples (with silence), and everythign is based on Estonian pronounciation system where a letter is pronounced same in any word (with some exceptions). Russian works so too (with some exceptions).

Jorge Nuno · Post by **Jorge Nuno** » Tue Feb 24, 2009 8:27 pm

Well this is a lot harder in EN/ES/FR/PT/others-I-know-nothing-about, because letters can have a TON of different phonetics, however the accents/tilde can help a little

HardWareMan · Post by **HardWareMan** » Wed Feb 25, 2009 3:40 am

I think you should use phonemes of syllables rather than letters. Then speech will be more clear. A long time ago I did the same thing for russian...

TmEE co.(TM) · Post by **TmEE co.(TM)** » Wed Feb 25, 2009 6:51 pm

I've got very good results using just letters, samples just have to be done right... I haven't done any work on this today :/

Snake · Post by **Snake** » Wed Feb 25, 2009 7:52 pm

HardWareMan wrote:I think you should use phonemes of syllables rather than letters. Then speech will be more clear. A long time ago I did the same thing for russian...

I've done this type of thing myself.

One thing I've always meant to do, but never seem to get around to it, is to attempt to extract samples from speeches by famous people... Could be fun

powerofrecall · Post by **powerofrecall** » Fri Apr 17, 2009 10:25 pm

What about rsynth?

http://www.speech.cs.cmu.edu/comp.speec ... synth.html

It's based on some really old, bad public domain code but it's marginally understandable and it sounds cool, haha

Chilly Willy · Post by **Chilly Willy** » Sat Apr 18, 2009 8:21 am

powerofrecall wrote:What about rsynth?

http://www.speech.cs.cmu.edu/comp.speec ... synth.html

It's based on some really old, bad public domain code but it's marginally understandable and it sounds cool, haha

It use floats for it's computations. It would be a bit of work to convert to integer math. It's also not likely to be anything close to "real time".