A ROBOTIC composer that can 鈥渨rite鈥� its own music is performing before a
critical audience in Zurich this week鈥攂ut if they don鈥檛 like what they
hear, they have only themselves to blame.
Called RoBoser, the system uses a neural network programme that makes music
using visual, tactile and audio sensors trained on the audience. As they
shuffle, cough and fidget, they鈥攁nd even the colours of their
clothing鈥攄etermine how the music will sound. Some 130 audience variables
affect the rhythm, melody and key of the music produced.
But the audience aren鈥檛 entirely responsible for the music, says Mark
Blanchard, who developed the system with Paul Verschure at the Institute of
Neuroinformatics in Zurich, and J么natas Manzolli at the State University
of Campinas (Unicamp) in Brazil. To prevent a cacophony, he says, the outputs
from the neural network feed a second system that provides musical templates for
the piece, with styles ranging from techno to jazz.
Advertisement
Verschure says that being inspired by its environment makes RoBoser similar
to people. 鈥淚 would claim that we have passed the musical Turing test,鈥� he says.