On Tue, 2011-10-11 at 10:26 +0200, Kassen wrote:
>
>
>
> No! But if you have a population, you might be able to build a
> predictor based on the whole lot...
>
>
> Right, yes, that could work. We could also analyse "families" that way
> to pick a part of the population where we expect individuals to be
> more interesting.
>
> Also nice would be trying to say something about the relationship
> between themes. If we somehow "breed" a theme that we like we will
> then need other themes to add variation to the song. At that point we
> need a theme B that doesn't only fit our criteria for "musicality",
> but also some criteria for being related to theme A.
>
> Oh, really quite exciting, this. I hope Dave will post recordings of
> what comes out.
I will - it's a bit unusual in that this whole setup requires a gameboy
at the end to hear anything, but I was also doing this with fluxa -
where the GP would be able to generate synth graphs as well. I was
actually doing this a while ago to automatically search for programs
which crashed the synth.
The code for the equivalent pattern program generator in scheme is here:
https://gitorious.org/flotsam/flotsam/trees/master/gpz
Growing scheme code is quite easy too, but a different technique to this
assembler stuff.
cheers,
dave
Received on Wed Oct 12 2011 - 06:37:58 BST