hi.

do you think its possible to separate vocals from syllables in speech (realtime)?
or - basically just get parts from the speech-flow which a speaker is able to hold out for a long time (like S sound, M, but not K or T)

thanks, krisztián

not really sure, but maybe you could do something like this with [bonk~]. but i think this ain't trivial.

Phase vocoder was originally created by AT&T to encodify speach into parts. Today we use it for time-stretching independent of pitch-shifting. I think there might even be a phase vocoder example in the PD documentation.

Not sure if or how you could use that to do what you want though.

Ha, I think I invented a word.

encodify = encode

I think what you're talking about is formant speech synthesis (analysis). I had to make a presentation about that, and honestly, I did without understanding a single concept about it! - it was the way beyond my audio and math knowledge. Here is the wiki page for it might be a starting point for you, and good luck!
http://en.wikipedia.org/wiki/Speech_synthesis

I don't know how you can separate sounds that can continue from instantaneous ones, but I have heard a Max patch that listened to speech and played back only the consonants. Basically it did an FFT and played back the noisier segments (maybe the I09.sheep.from.goats.pd help patch would be useful here).

I typed up a whole thing that I then realized had nothing to do with your question.

I think Ichabod's suggestion is on the right track.