16.6 Musical examples#
Holly Herndon and Jlin - Godmother (2018)#
Composer Holly Herndon, her partner Mathew Dryhurst, and developer Jules LaPlace built an AI system they named Spawn. To create Godmother, Spawn listened to the music of the producer Jlin, Spawn’s “godmother”, and reimagined it in a model of Herndon’s voice. Herndon states that the track was generated from silence without samples, edits, or overdubs. Godmother later appeared on Herndon’s 2019 album PROTO, made in collaboration with Spawn. It is an early example of an artist treating a voice model as both an instrument and a collaborator.
DADABOTS - Relentless Doppelganger (2019)#
DADABOTS, the duo of CJ Carr and Zack Zukowski, train neural networks on audio recordings, and then let them generate new music in its style [CZ18]. Their models are based on SampleRNN [MKG+17], a language model over audio samples, much like WaveNet. Relentless Doppelganger is a model trained on death metal that generates music 24/7 live on YouTube. The stream began in 2019, and has been playing around the clock mostly uninterrupted for over seven years.