by Michael Mayr (AT)
A neural network trained on a dataset of selected Techno tracks generates samples from an array of 64 parameter values. Each of the parameters and their combinations modify sound in a unique way, activating and amplifying different structures and patterns learned by the neural network. Due to the black box nature of the deep network, it is very hard to describe and understand what each parameter does and how all 64 of them are interconnected. Shaping and finding the right sound remains a trial and error process, however with each rendered sample retaining enough rhythmic and spectral quality from the training data, there almost is no 'wrong' sound to find.
Tags: Performance
  • 12 September

    SOUND CAMPUS: Michael Mayr: TechnoGAN

    18:00 - 18:45 · Hauptplatz 6, Courtyard