Brief description: Synthesize new music by running visual/textual representation of music through visual/language deep learning models
Two possible pathways to experiment with:
a. train styleGAN/style transfer on datasets of sheet music, and create new musical notation – try to see if what comes out is playable
b. train GPT-2 on datasets of music represented as MIDI notes, create new music through it, and play it.
To make the datasets more consistent I will focus on the works of one classical composer.
I have been interested in the music of Richard Wagner and have worked on applying music from Wagner’s operas to new text in a different language. This work was done by mere humans, no machines were involved.
http://www.gilsperling.com/a-yiddishe-wagner/
There is something very reductive about the Style Transfer models for painters – there’s something that feels fitting about applying that reductivism to Wagner’s music, a man who contributed more than anyone to the myth of the unique genius composer.
Posted in Spring '20 - Intro to Synthetic Media |