"genetic engineering of synthesis patches". ok, that sounds like a potentially nice PhD.
I'd wonder if the genetic algorithm or (as I studied a bit in the past,but never really worked on) the cohonen network learning from (which ?) examples can do simple thing like counting to 4 binary, or distinguishing a violin timbre at central C from a piano note of 442 HZ ? I mean, otherwise it may be interesting, but only in the sense of slightly funny, for a very short time.
When I compared DX-7 (and some other) paraeter spaces back in the 80s, there was a program on the Atari ST computer which would do harmonic analysis, but I didn't have that. I made a C program plus nice interface that would search over thousands of patches for sounds with parameter-sets closer than certain sum of the absolute value distance of the parameters, and wouls check if the sounds would end in the similar categories, or were audio-wise indeed as similar as the distance would dictate, which was more than a bit the case, so it was a useful way to sort synth sounds.
Of course, for people who know a bit about FM and it's possible algorithms, this wasn't necessarily true, since the wildly changing spectrea resulting from a slightly different algorithm parameter (from 1 to 32) would suggest otherwise.
Of course this leads us to a question of the size of the range of the set of possible sound-patches. Luckily for FM this was a large space, with lots of interesting and varied sounds. it is likely that the actual range-size of rougly 2 to the power of 150 bits, has a lot of small variation possibility around "sensible" sounding patches, and probably a lot of unused space not covered by the few thousand large database of them.
The general question of what the range-filling of a synthesis algorithm is given all the possible sounds of a few secnds in CD quality, even projecting to a result where obvious affine type of transformations are filtered out (volume variations and time shifts and scalings), like a fraction of 2*(2^16)^(2*44100). Probably, if we take into account all the great records (CDs) and wonderful sounding instruments and syntehsizers, we'd better put on a CD instead of trying out some artificial genetic character and hope for succes at athat level.
Darn, I'm starting to sound picky-er than mr. Sheldon himself (from the BBT show)... :) Maybe i do need that CALtec job!
Theo V. -- dupswapdrop -- the music-dsp mailing list and website: subscription info, FAQ, source code archive, list archive, book reviews, dsp links http://music.columbia.edu/cmc/music-dsp http://music.columbia.edu/mailman/listinfo/music-dsp
