Databots is a band powered by deep learning software developed by CJ Carr and Zack Zukowski who met during their time at the prestigious Berklee College of Music in Boston, Massachusetts.
Databots, which is actually an AI generating its own renditions of what death metal sounds like, was based on the Canadian extreme metal band, Archspire. The creators wanted to prove a neural network was capable of capturing subtle differences between death metal, math rock, and other underground genres of metal.
It’s based on a recurrent neural network-computing architecture that learns patterns within a large amount of user data in order to predict which elements and patterns are most common and then it recreates them.
The duo explained the process in a 2017 paper posted to the Cornell University arXiv server. It starts by feeding segments of music to the AI model, a few seconds at a time. As the process moves along, the AI learns to identify features and uses the learnings to produce more detailed samples, including main song riffs and transitional phrases.
Aside from the 24 hour YouTube stream created to showcase the project, the team has gone on to release ten additional albums based on the sounds from various bands like Aepoch, Meshuggah, and Battles. For each project, they have Databots analyze subsets of an artist’s discography and work from it to create new music.
Originally posted by Vice.com: https://motherboard.vice.com/en_us/article/xwnzm7/this-youtube-channel-streams-ai-generated-black-metal-247