Stop searching for parameters today, your true love is only one training stage away!
but seriously… this uses the mlpclassifier~ to associate hate/love labels to parameter combinations using mfcc statistics as the inputs. You can then refeed (albeit a little manually) predictions back in if you think they are correct to help hone the network.
That being said, mucho points for using the Fourses algorithm as the synth.
I love it but in the end I think the Tinder model is too binary for me. I think I’ll need more classes (dirty love, clean love, boring, shit) and that might help me. Or maybe just 2 sliders to a regressor (dirty-clean and love-hate) and then predict random options for me…
more soon!
Oh yeah the slider is a good idea. I think it is so easy to mistrain it once you get into a higher order of inputs. I tried doing it with 3, ala this but I ended up forgetting what I previously did, and I don’t trust my intuition to be that consistent in the moment.
2 sliders might account for my diversity of tastes 
