Transforming musical audio data into visual art : a study of machine learning methods to simulate Wassily Kandinsky’s process of creating visual art from sounds

Abstract

Data is used in nearly all aspects of life, and is becoming more and more advanced. There are machine learning models and algorithms used in advancing human life, in modifying behaviour, and protecting data from harm. This research is focused on finding out how to utilise data in advancing something completely human; art.

Data has been traditionally connected to parts of the professional world where we can discern facts. However, applying data to art is immensely more difficult because art can’t necessarily be categorised into specific facts. Its very nature is highly subjective to the artist’s vision and talent, but also to the perception of its audience.

To understand how to use musical data to create visual art, we have researched ways to use music and art theory in combination with machine learning methods. This research studies audio source separation and feature extraction, and utilises available audio data visualisation libraries.

To create a final visual composition, the results from these methods were combined with the colour and form theories of the abstract painter Wassily Kandinsky.