Emmett Williams' 'Cellar Song for Five Voices'

(~1960/2021), generative audiovisual work [duration variable]

08/07/21-31/07/21 - premiered online as part of VIRTUALLYREALITY SS21: The Long Triumph' (Manchester, UK).


experience the work here.


Screenshot 2021-07-08 at 13.55.45 1.png

In the 1960s, somewhat oracularly, Emmett Williams compiled a poem-come-dataset of systematically shuffled phrases that filtered prescient themes of ecology and nonlocality through the skies and subterranea, bluebirds and blackbirds, life and extinction. 


This new adaptation of 'Cellar Song for Five Voices' feeds this dataset through artificial intelligence and variable autoencoders to create new attenuations of the original poem and its performers. Text, deep-learned readers and their voices are systematically outputted at various epochs of the training process, generating a nexus of uncanny identities that reckon with AI’s interlockings with deep ecology.

video demonstrating the webpage:

example iterations:

(left to right: Rosalind Ridout, Callum Coomber, Izzie Mortimer, Lucy Nolan, myself)

webpage screengrab:

Screenshot 2021-07-16 at 15.22.50.png


Concept - MICHAELBRAILEY & Callum Coomber
Max programming - Michael Brailey

Machine learning - Laura Ibáñez
Sound/video input dataset - SHOAL​ (Callum Coomber, Izzie Mortimer, Lucy Nolan, Rosalind Ridout) and Michael Brailey