DEGEM CD 23
Daniel Bisig, Ephraim Wegner, Thomas Wenk
Technical sounds from analog media of the last century are processed by a machine-learning based approach called Deep Dream.
Deep Dream audio is an adaption by Daniel Bisig and Ephraim Wegner of the well known Deep Dream image processing method. The method employs a convolutional neural network that has been trained as a classifier on different types of acoustic events produced by tape recorders. This network has been used to create acoustic material by iteratively amplifying those acoustic properties in an initially noisy signal to which selected network layers respond with high activity.
The work focuses on the generative and power-guzzling process of generating these sounds accordingly to the Deep Dream Principle. For this reason no additional digital effects have applied to the source material and processed sounds.
http://www.e-wegner.net/
http://www.twenk-cassettology.de/
http://www.bitingbit.org/
