Recently I built a neural network with pytorch to generate 16khz audio samples based on a dataset of kick drums. I wanted to use AI generated audio samples in my music production and figured this might be a fun project to learn how different nn layers interact as well as understanding the mechanics behind generative applications of AI.

I was able to successfully train this network on my gaming pc with a 2080 GPU. Whilst the generated outputs resembled the input dataset, it doesn’t generate a super diverse set, I think it needs to be more sensitive to the input noise. Anyway it was really fun to experiement with this and I’ll certainly come back to it.