None by Magnus Bugge

A machine learning based approach to deep listening, atmosphere and fidelity

In ‘None’ I became occupied with the vast worlds of sound in latent and liminal spaces that would normally be considered ‘silent’. I recorded such sounds: room tones, barely audible field recordings, tape hiss, gear hum, and processed them in various ways to listen closer. I applied machine learning and trained a model on this desolate sound world. The sampleRNN model is ‘a novel model for unconditional audio generation based on generating one audio sample at a time’. One audio sample is basically the elementary particle in digital audio. It is a single point that is meaningless without the context of what becomes before and after, in itself it has no direction, and crucially, no sound. However, knowing the trajectory and velocity you may predict where it will travel based on an idea of what music is. In my case, the intelligence have never heard music before, only silence, and thus predicts based on this.

electronic music maker

magnusbugge.com