shadow (2018)saxophone, no-input mixer, and DMX lights
The sounds of the no-input mixer are analyzed in real-time using timbral descriptors, which are then sent to a neural network for classification into one of four categories: distorted noise, high squeal, low impulses, or quiet sustained noise. These classifications then trigger different audio-analysis-to-lighting-parameters mapping strategies.
Using both a categorical classification of sound and a raw parameter mapping system allows for the minor changes in the no-input-mixer’s sound to create minor changes in the lights, while also reflecting the fast and drastic changes in sound with drastic changes in the lights. This system is also used in my work, feed.
I’ve presented on this work and the neural network classification system in a few presentations.
tags: feedback , machine learning , lights