Julian Skar

Hi all, im working with a piece for voice and real time electronics.

Trying to build a patch that easy can tell difference between "percussive" sounds like noisy phonetics from wovels and pitch. I have tried a lot of libraries allready and different objects, and just cant make something stable. Its really important be cause this is supposed to be the "heart of the patch" where the singer can trigger random ambisoncs points, preset changing, coll objects etc etc.

I would really appreciate if anyone had a concrete suggestion, more than a hint for libraries etc be cause time is running out (deadline for my piece was last week) and i just cant crack this. Typical problem this, where i as a composer can have many ideas and try to dive into the world of interactivity and electronics, but at the other hand have to struggle with lack of solid DSP background, and often basic processing that´s probably very easy for sound techs and engineers.

As i said any help would be very appreciated, and hopefully i can contribute with some knowledge about music or maxmsp help for others that are even newer to this world than i am next time.

Trying to build a patch that easy can tell difference between "percussive" sounds like noisy phonetics from wovels and pitch. I have tried a lot of libraries allready and different objects, and just cant make  something stable. Its really important be cause this is supposed to be the "heart of the patch" where the singer can trigger random ambisoncs points, preset changing, coll objects etc etc.

help-to-seperate-pitch-from-consonants

What you are trying to do is indeed not trivial and quite complex. However, I know of someone doing a similar work with great documentation of the process. Have a look at this page 

http://orchestraofspeech.com/tag/ftmmubu/

It is a bit technical and uses some of the free IRCAM libraries which have very poor documentation.

 What you are trying to do is indeed not trivial and quite complex. However, I know of someone doing a similar work with great documentation of the process. Have a look at this page http://orchestraofspeech.com/tag/ftmmubu/

It is a bit technical and uses some of the free IRCAM libraries which have very poor documentation. 

You can, of course, make a primitive estimate with the xerox~ object. Lots of zero crossings in a short amount of time is, in general, indicative of noise.

It's very rough and ready, but works well for some purposes. There's an example vocoder patch showing it in action.

for speech, an FFT-based transition analysis works very well.

the attached max4 patch was made after some pd example from 1995 (settel?)

it splits the incoming signal into transients and the more harmonic rest of the signal.

you can now analyse the rms power of these two outputs and distribute the input signal accordingly onto 2 channels to seperate the transients from the rest, just like you would do it when building a simple dynamic gate (for noise supression or so)

if you make it a pfft, you can use fftinfo~ to find the delay time which is right now just a messagebox. :)

analyzer~, found at cnmat.berkeley.edu, has brightness or noisiness which might help.

Or maybe something from the zsa.descriptors: 

http://www.e--j.com/index.php/what-is-zsa-descriptors/

Or maybe something from the zsa.descriptors: http://www.e--j.com/index.php/what-is-zsa-descriptors/


help to seperate pitch from consonants