Participants 2022
Dutillove
AI Song Contest 2022 / participantsTEAM / Dutillove
SONG / AI Sketch No. 1
TEAM MEMBERS / Berker Banar and Cenk Esen
About the TEAM
Berker Banar – Studied Electrical and Electronics Engineering at Bilkent University, Ankara, Turkey, and Electronic Production and Design at Berklee College of Music, Boston, MA. Currently, pursuing his PhD in Computer Science focusing on “Towards Composing Contemporary Classical Music using Generative Deep Learning” at AIM CDT, C4DM, Queen Mary University of London, under supervision of Simon Colton.
Cenk Esen – Born in Istanbul, Turkey 1999 Son of musicians Randy Esen & Aydin Esen. Started playing piano at the age of 9, seriously at age 16, after participating in Berklee College of Music’s “Umbria Jazz Clinics” program in Perugia, Italy. Studies Piano Performance at Berklee College of Music and was also a part of the Berklee Global Jazz Institute, founded and led by master Danilo Perez.. Now based in Europe.
Berker and Cenk have been playing together in various occasions ranging from an improvisation duo to larger ensembles.
About the SONG
AI Sketch No. 1 was composed by a generative deep learning model, which we specifically tailored for the 20th century music. The piece was arranged by Berker and Cenk with minimal touches and converted into the sonic domain. Our goal was to create a contemporary sound with an open and breathing flow, yet with a medium-paced journey ending with a not-so-stable closure. We hope that the listeners will focus on the contemporary material that our AI model has generated and also the note selections, themes and the expression. Also, our hope is that this piece will contribute to bringing more attention to the conjunction of contemporary classical music and generative deep learning.
About the HUMAN-AI PROCESS
GPT-2 is originally a language model created by OpenAI. We repurposed it to generate MIDI music in this context by further training it with a subset of GiantMIDI-Piano music dataset, where pieces from 1900-1950 period were specifically selected. All the themes in this piece were composed by the GPT-2 model and selected among hundreds of candidates using some musical metrics. Technical development of the generative system was done as part of Berker’s PhD research at QMUL under supervision of Simon Colton. The selected themes were then arranged by Berker and Cenk, and the themes were mostly used as they were besides a few touches. Also, Cenk played 15 seconds of MIDI piano aligned with the arrangement of the piece and MIDI music composition was converted into the sonic domain using ROLI’s Feedbacker Piano patch by Cenk as well as some audio processing.
This piece was intended to follow contemporary aesthetics and the tools used were kept to a minimum as part of the artistic practice, which was found to be focused and inspiring. As part of our experience, we’ve been surprised by the contemporary sound offered by the AI model, which was in the same direction as our goals for this piece.
We would like to thank Jesse Engel for his time and involvement as part of the Mentorship Program.