Relentlessly Infinite: AI Generated Death Metal Now Streaming
Death metal is not a genre everyone finds enjoyment in, but I think we can all stop to appreciate Dadabots. Developed by musicians and machine learning researchers CJ Carr and Zack Zukowski, Dadabots is an AI band, trained via a recurrent neural network, dedicated to creating an endless stream of “Black Metal and Math Rock”.
The entire concept is based on the work from their 2017 paper: “Generating Black Metal and Math Rock: Beyond Bach, Beethoven, and Beatles.” The Dadabots stream was trained using the music of Archspire, a fast-paced technical death metal band from Vancouver, British Columbia. According to Carr, in a conversation with 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.” In that same conversation Carr reveals that the Dadabots stream is entirely autonomous, “running on a Linux server somewhere in South Carolina.”
The stream being produced by Dadabots manages to sound like highly technical death metal, and completely inhuman at the same time. The speeds at which the drums and guitars are playing are unachievable by most humanoid musicians, and the vocals are total guttural gibberish. I’ve had the stream running in the background of my work day for the better part of two hours now, and at first it felt too jarring and mechanical, but after a few tracks (?), I’ve been fully converted as a fan of the Dadabots Archspire model.
Dadabots also has a robust Bandcamp account. It features 10 albums that have been generated via the same training process, and include artists ranging from The Beatles to NOFX to Meshuggah. I am making my way through these albums and chose to start with “Bot Prownies,” generated from a model trained on NOFX’s “Punk in Drublic.”
Follow Carr and Zukowski’s efforts toward exploring “style-specific generative music” from more what they call “genre outliers,” and check-out their paper for a more in-depth look at Dadabots and information on how the models are trained.