Nolan Lem

neural ordinance (2016)

neural ordinance is comprised of sounds that are a result of my computer being trained to produce industrial noises. In this type of deep learning, recurrent neural nets literally teach the computer how to produce sounds that are representative of machines themselves. As such, this piece focuses on a large corpus of field-recorded sounds that include audio related to industrial drones, server farms, consumer electronics, HVAC noise, etc. After processing these recordings, the computer ‘dreams up’ sound based off of its own idea of what industrial noise is. If we can treat the computer as a superlative machine, the neural network seeks to reify a sonic representation of what the computer itself thinks it sounds like. In this way, it shows the computer trying to listen to itself.

In this instance of the piece, the noise emanating from the speakers on the CCRMA stage were included into some of the training sets used in the synthesis. As a result, the output sound is a mixture of both real-life analog noise and the computer’s interpretation of the same. The sounds undulate, swell, and breathe to form an ecology of machine-interpreted awareness, one that suggests a strange convergence of the real and the digitally imagined, the sentient and the synthetic.

The title is taken from the term 'noise ordinance' which refers to the noise regulations that are typically enforced by city zoning codes. In this case, the neural network acts as a governing agency that imposes its own definition of what is constituted by 'noise'.
stereo version
download original 8 channels
channels correspond to speakers in parallel-pairs format (see diagram below - 1 corresponds to 1.wav, etc.)