Talking Heads

A Sound Generative Project based on the imperfections of AI

Featured in Ircam Forum 2022

With Talking Heads, I am exploring my interest in how to create generative work by emphasizing the glitches caused by the imperfections in contemporary AI technology. The original idea for Talking Heads came from a p5.speech program I created. The program is a simple loop for AI to listen to itself speak, analyze and speak again based on what it thinks it hears. I became addicted to this loop because the loop would generate a lot of amusing glitches. After 1 minute or so, the AI starts to fight itself. It’s like the telephone game.

Another interesting element is that during the AI loop, any sound from the outside will affect how it analyzes what it hears, and it will impact the intensity and direction of the glitch.

Based on this idea, I created Talking Heads to expand on this concept of a generative glitch. I built a Max patch that analyzes the p5 AI voice it hears coming from the computer speaker and generates a sound piece that can only be heard within the headphones, creating two distinct sonic environments the listener/viewer can experience when encountering the work. The patch also uses the same analyzed data to control the behavior of a 3D visualization projected in front of the viewer/listener.