[sdiy] For you FM synthesis fans

[Scott Nordlund]

> On Mon, 17 Nov 2014, Ben Lincoln wrote:
>
>> There was a Sound Diver module for the K5000 (produced by Kawaii, IIRC)
>> that would allow for most of that sort of thing. It would also do a
>> Fourier transform on a WAV file and attempt to recreate it using the
>> K5000's synthesis model.
>
> I vaguely recall that back in the times there at least was *talking*
> about a software which would do a similar thing on the DX7, i.e.
> grasping the characteristics of a sound and creating a close mapping to
> a DX7 FM setting.
>
>Rainer

This is much more difficult than resynthesis for a simple harmonic additive synth, because the spectrum resulting from a given set of FM parameters is far more unpredictable. It's easy enough with only two operators, but not when you consider the full gamut of possibilities, with chains of operators, feedback, operator detuning, etc. There's a paper by Matthew Yee-King (who incidentally released an album on Rephlex) describing a genetic algorithm that will automatically program any soft synth to match, as closely as possible, an arbitrary sound. It was tested on a 4 operator FM synth. The results are better than a human programmer can manage, though it's limited because it's only analyzing a single sound and thus won't program expressive velocity sensitivity or keyboard scaling curves. It starts to become too computationally intensive if there are many more than 4 operators.

Here's the paper: http://yeeking.net/pdf/roth_yee-king_ICMC2008.pdf
And a further one that I haven't read: http://yeeking.net/pdf/AES_paper.pdf

There are unsubstantiated rumors of Yamaha using software like this to program the presets for their FM synths, but to me this seems kind of absurd.