Training new A.I. model

in #music4 years ago

I'm trying to evolve the music composer A.I. as much as possible.

While finally I've decided to use the magenta tensorflow script instead of a more basic one, I realized that with the current dataset the model can not produce any more better results.

I'm going to publish the best music the model created in few days, but now it is time to try to use a more sophisticated and complex dataset and see what is going to happen.

So now I started already to train with Chopen and Debussy dataset, something immensly more complex than the previous one.

chop.png

It is very possible that the dataset is simply too complex and than maybe I did some mistakes, but I really hope the outputs will make at least a bit of sense.

After this I'll try to train with modern dace music if I'll be able to find few midi files on the net.