A quick search for “timbral descriptors” turned up a working paper from IRCAM that I think is pretty useful: http://recherche.ircam.fr/projects/cuidado/wg/management/deliverables/IRCAM_WP215_M2.pdf. The summary lists the following parameters as the most salient: spectral centroid, the spectral spread, the spectral deviation, the effective duration and attack time and finally roughness and fluctuation strength. The paper also mentions that for harmonic sounds, F0s must be set to be identical before comparisons can be made. It looks like someone at IRCAM used the results of this paper to put together a web application that computes timbral distances. See http://recherche.ircam.fr/equipes/cuidado/Timbre.php3.

I won’t have time to experiment with your code for a while, but when I do, I’ll let you know if I try to integrate some of these findings into the fitness function.

Yes, I’ll be investigating “smarter” fitness functions. Since the system is modular, it’s easy to switch the fitness function, even on the fly for multistage GA. Fitness funcs. based on psychoacoustics are especially interesting because they will most likely converge to perceptually better sounds in smaller time. It will probably narrow the search space considerably, yielding to more satisfactory and interesting results.

And for anyone reading this, I could use any help.

My only suggestion would be using a more sophisticated fitness function, one based on psychoacoustics. For instance, we know that the attack part of a sound has a disproportionate influence on its perceived character; a fitness function could overweight the attack to reflect this insight. Please keep blogging and sharing your work!

Cheers!

CpILL, as I’ve mentioned above, the example code I’ve used in the video is included in the documentation (GAPmatch help). The fitness function is tricky, it uses an UGen I’ve coded in C for the purpose. It compares the MSE distance between complex spectrums of the source and synthesized sound and the MSE distance becomes the similarity rating therefore the output of the fitness function (lower is better). I’ll let you know when the paper describing the internals is published.

Here is a paper by Ricardo A. Garcia where he grows sound synthesis graphs (instead of parameters to fixed synthesizer graphs), I use the same similarity rating formula, but discard the phase:
http://galileo.cincom.unical.it/Music/workshop/articoli/garcia.pdf

And very nice video demonstration!

Pretty impressive!

Thank you Batuhan!