digital sound

Endel

Dadabots—a fake band powered by deep learning software—was developed by CJ Carr and Zack Zukowski, two musicians and technologists who met while they were going to Berklee College of Music in Boston they told The Outline. It’s based on a recurrent neural network—computing architecture that “learns” patterns in a large amount of input data (in this case, death metal) in order to predict what musical elements and sequences are most common and recreates them.

They broke down their process in a 2017 paper posted to the arXiv preprint server. They start by feeding the AI model short segments of music, a few seconds at a time. As this training goes on, the AI learns the identifying features and starts to produce more and more detailed samples, including riffs and sectional transitions.

In the paper, Carr and Zukowski wrote that initially, they were surprised by the result.

“While we set out to achieve a realistic recreation of the original data, we were delighted by the aesthetic merit of its imperfections,” they wrote. “Solo vocalists become a lush choir of ghostly voices, rock bands become crunchy cubist-jazz, and cross-breeds of multiple recordings become a surrealist chimera of sound.”

While it doesn’t sound totally human—because the vocals in each track are distorted gibberish, notes are held without room for breaths, and some of the guitar riffs are at speeds most people couldn’t achieve—the general feel and instrumentals are convincing, especially to the untrained ear.

With Dadabots, besides the YouTube stream, they have released 10 different albums based on the music of metal and experimental groups like Aepoch, Battles and Meshuggah. They then curate the best-sounding tracks into an album. They characterize their work with Dadabots as working towards “eliminating humans from black metal.”

For each project, they have Dadabots analyze "subsets of a single artist’s discography,” and work off of it to create its own work. The music that the livestream is based on is a Vancouver-based technical death metal band called Archspire. And it produced another pleasant surprise.

“Most nets we trained made shitty music. Music soup. The songs would destabilize and fall apart. This one was special though,” CJ Carr told Motherboard. The Archspire Dadabots created much more consistent, stable music. Carr’s guess is that because Archspire’s music is played at such a high tempo, it stabilizes what the bot puts out. “It's autonomous, running on a linux server somewhere in South Carolina,” he said. “You're hearing everything it makes.”

In the future Carr and Zukowski hope to include some kind of audience interaction with Dadabots.

This version of Jasmine by Jai Paul is created using the BRONZE AI engine. On each listen, Bronze performs a unique and infinite playback of the piece.