learning to die
learning to die
for Alto-Voice, Doublebass-Clarinet and neural network synthesized sound
Duration: ca. 9 mins.
The nine short pieces are intended to show how a neural network learns to generalize a data set. The title of these miniatures - learning to die - refers to the training data used, famous death arias from opera history (a character dies on stage). The audio tracks are completely synthesized by the model and range from noisy, less gestural results of the first training epochs to a (somehow) vocal-orchestral performance of the fully trained model.
commissioned by and written for Helena Sorokina und Marco Sala
Alte Schmiede Wien, 21.02.24