KVR Audio

studiowaveform · Post by **studiowaveform** » Tue Dec 26, 2017 9:37 am

I was wondering if this is possible, and is it the method that programs like melodyne use for polyphonic removal?

BertKoor · Post by **BertKoor** » Tue Dec 26, 2017 9:56 pm

You mean: is there an automatic way to determine the frequencies and arrangement of all operators on a FM patch?

Nope. The really interesting FM patches result in total chaos.
With quantum computing maybe!

Afaik Melodyne uses regular FFT and looks for correlations between bands. FM modulators create harmonics and these show up at different frequencies than their own.

studiowaveform · Post by **studiowaveform** » Wed Dec 27, 2017 11:02 am

Yes, I meant extract operators from an FM patch. Two for my purposes.

Thanks for the clarification

stratum · Post by **stratum** » Wed Dec 27, 2017 1:04 pm

BertKoor wrote:You mean: is there an automatic way to determine the frequencies and arrangement of all operators on a FM patch?

Nope. The really interesting FM patches result in total chaos.
With quantum computing maybe!

Perhaps this would work for the less interesting cases?
http://wwweic.eri.u-tokyo.ac.jp/compute ... FD0AEB.htm

Smashed Transistors · Thu Dec 28, 2017 11:20 am

If you want to find synthesis parameters for a two operator FM tone, you can have a look at papers by Horner and/or Beauchamp such as "Genetic algorithms and their application to FM matching synthesis"

Music Engineer · Post by **Music Engineer** » Thu Dec 28, 2017 12:38 pm

studiowaveform wrote:Yes, I meant extract operators from an FM patch. Two for my purposes.

if it's just two operators (in feedforward connection, i.e. just a carrier and a modulator) and both, carrier and modulator are sinusoidal, there are certain rules - like the frequency ratio of carrier and modulator (c/m ratio) determines, where the frequency components end up and the fm-index (modulation strength) determines their amplitudes via bessel-functions

. i guess, in this simple two-operator case, one may look at the spectrum and use the frequencies to reverse engineer the c/m ratio and their amplitudes to get the fm index. this is the classical paper that describes these rules:

https://ccrma.stanford.edu/sites/defaul ... aper-2.pdf

edit: but then, i guess with just thes two parameters (c/m ratio and fm-index), one could also just go with trial and error. ...and for more complex fm-setups, it gets really complicated real quick

Possible to extract the musical notes from a single FM tone?