GANKYOKU

‘GANkyoku’ is the name given to any pieces created by my Generative Adversarial Network with the same name. These are the first full-length Contemporary Classical pieces to be entirely composed by a deep neural network. The pieces were commissioned by Shawn Renzoh Head, following my award in his 2017 composition competition.

The aim with these works is that they should maintain key aesthetic qualities expected of shakuhachi music, derived from the traditional repertoire, or Honkyoku. Rather than purely creating pastiches of classical shakuhachi music, I sought an AI model that adds novel features and idiosyncrasies in a somewhat systemic way, making worthwhile contributions to the contemporary shakuhachi repertoire.

The three pieces were all composed by the same generative AI model, despite their great variances in length and material, suggesting the model has successfully learned many of the deeper underlying features of the classical solo shakuhachi repertoire, allowing it to create original and wide-ranging works with clear referral to the Honkyoku aesthetic. These pieces represent our three favourites out of approximately 30 generated candidates.

A paper concerning the development of the GAN used to compose these pieces has been accepted for publication in the ICMC-NYCEMF 2019 proceedings, and can be read here.

Video recording of ‘GANkyoku’ I, II and III and Girton College Chapel, University of Cambridge, 26th January 2019. Video recording by Bilal Hasna, edited by Shawn Head. Audio recorded and mixed by Omar Peracha.

Performer:

• Shakuhachi - Shawn Renzoh Head