ABOUT THE TEAM

All band members are or have been affiliated with the Music, Science and Technology Research Cluster in the Department of Music at University of York, UK. Tom is a Lecturer (Assistant Professor) in Music Technology, and Co-founder of Music Artificial Intelligence Algorithms, Inc. Liam is an Associate Lecturer with specialisms in sound recording and music production. Lynette is an audiovisual maker from Singapore and recently graduated PhD student from University of York. Mark, Jemily, and Alex are current PhD students. Mark is a saxophonist researching neural networks in the waveform domain. Jemily researches immersive and interactive podcasting, and is a musician and songwriter in her own right. Alex works on neural networks applied to symbolic music representations.

ABOUT THE SONG

The starting materials for Circus were all generated by AI, including our band name The elephants and the, which gave Jemily the idea to use "circus" as a keyword for the lyrics generation algorithm in the first place. Melodies, chords, baselines, and drum tracks are from the MAIA Markov algorithm, and initial lyrics are from theselyricsdonotexist. Novel aspects of AI this year include the first ever tonal saxophone riff and solo generated in the waveform domain, an algorithm to suggest alternative lyrics based on clips of instrumental audio, and a video also generated largely by AI. The six humans involved in this collaboration had lots of fun in the process, so we hope you enjoy listening and dancing to the result this summer. Let the circus come to town!

ABOUT THE HUMAN-AI CO-CREATION PROCESS

To our knowledge, Circus is the world’s first song with tonal AI riff and solo generated in the waveform domain. We trained a generative adversarial neural network (GAN) on a dataset of saxophone audio recorded by Mark, and used that network to generate sax audio for the song, which we think sounds awesome and novel!

We used the MAIA Markov algorithm to generate melodies, chords, basslines, and drum beats. Lyrics were generated from https://theselyricsdonotexist.com and an in-house neural net developed by Alex that predicts lyrical content given instrumental audio. Lynette explored the StyleGAN algorithm and the Generative Engine by RunwayML to generate image and video materials to go with the song. Additional manipulation to the images were done in Processing, with post-editing for further enhancements. Initially, Jemily explored the lyrics and MAIA Markov material, and devised four clips. We then met as a group for 90 mins on Thu 29th Apr to consider all AI materials and Jemily’s clips, and in that meeting devised the chorus. Jemily and Tom had a subsequent 2-hr songwriting session where we combined more MAIA Markov melodies and lyrics to arrive at a verse, pre-chorus, and middle 8. Jemily and Tom both worked further on their own versions (Jemily’s quite jazzy; Tom’s quite cheesy), and by Mon 3rd May shared those song drafts with Liam to undertake music production activities.