posted 08 April 2010 08:51            

---

hi
i'm davide, from italy. I'm just starting learning kyma: so deep and endlessy creative!
and so frustrating!
yesterday i was looking to perform a spectral averaging of a long sound file with kyma (and then go crazy with other processing), something i used to do with soundmagic spectral plugin... and i got lost. is there a sound object that can already do that (i didn't find anything), or anybody can point me to a starting sound to be modified?
thank you in advance
davide

IP: Logged

posted 08 April 2010 09:57            

---

As I dont know the app you're mentioning, I'm not sure I am 100% clear on what you want a "spectral averaging" to do?

You can set the partial frequencies of a spectrum analysis file to their average frequencies across the entire file (described in KYMA X Revelead), but this does not necessarily really do what you want it to, as for mixed/complex sounds the result will not necessarily be the *average spectrum* in terms of timbre but the sum of partials running at their *average frequencies*. This may sound very different to what you actually think that it might sound like. It'll probably still sound pretty cool in most cases, of course...
Maybe you're more looking for spectral snapshots of the file in question, added together and the amplitudes then divided by the number of snapshots (basically "accumulated frames"). This would not change the frequencies of the contained partials, rather *all* frequencies contained would be present in the outfile, but scaled to a level relative to the duration they are actually present in the source file. I wouldn't know how to do that OTOH, that would probably be easier to do with an FFT/iFFT combo than with a spectrum file; or you may need to write a tool for that.

IP: Logged

posted 08 April 2010 11:40            

---

hi thank you for your answer.
i was thinking something like an average of the spectral information over n windows. you get something like a "blur" effect, but the harmonic content is preserved.

IP: Logged

posted 08 April 2010 12:05            

---

OK. I would try something like an FFT into n parallel delays, each of which delays by one (times number of delay) FFT buffer lengths (so you get direct + (delay by 1 buffer) + (delay by 2 buffers) + (delay by n buffers). Then sum the delays, do an iFFT, divide level by n, done. It will then take buffer*n samples before the output is really doing what it is supposed to do.

At least I think it *should* work like that. I won't be near my Kyma before Monday, so I can't verify that I'm actually giving good advice ;-)

You could probably generate the parallel delay lines with a replicator module instead of wiring them all by hand.

Keep in mind that the FFT module outputs two channels for 1 input channel: amplitudes and phases.

Anyone with a better idea (I'm not really a veteran myself ;-) ) ?

IP: Logged

posted 08 April 2010 15:07            

---

Have you tried replacing the frequencies with their average in the Spectrum Editor? Or perhaps adding smoothing to the frequency envelopes in the editor (rather than completely replacing the frequency envelopes with the average?) Not sure whether this is the effect you are looking for but it might be interesting in any case.

IP: Logged

posted 17 April 2010 06:41            

---

In my experience I found that for getting more flexibility on realtime manipulations/interpretations of spectral data its most interesting to design an oscillator bank with your own resynthesis and treatments , by using replicator and voice conditional track selectors. That way you can have track based smearing using filters or delays with feeback and so on.

IP: Logged

All times are CT (US)