Hi, I finished a small part of the project but I am now taking some time off until it again has a lot of my passion.
I am not creating an algorithm that writes music. I try to use Darwinian Evolution. The input will be massive and very varied in the nature of mutations. So every generation I create many different distinct versions. I also use gender to recombine the music genes every generation.
So each generation I create many different versions from the original 'musical DNA', which will code for a single note at the start. So each time there is a mutation that adds a new random note through pure chance I will select that mutation out of many. Then I use the selected versions for a new generation.
Big problem is the amount of selection that has to be done. It has to be done by humans. Humans need to select the mutations through their taste for music using 'books, feelings, childhood, happiness, other music etc.', like you decribe.
That way they can design the music indirectly through Darwinian evolution. If this were technically possible the results may be very interesting.