Modeling Audio Flows and the Geometry of Information

New paradigms for audio signal processing and representation

A digital audio signal contains multiple flows of high-level information such as notes, instruments, and the musical form. The analysis and extraction of these contents is a recurrent problem in science and computer-music technologies that attempts to bring these two aspects of sound—symbolic and identifying—closer together.

The objective of this project is to use tools recently created for the geometry of information to suggest new paradigms for audio signal processing and representation.

The geometry of information is a recent mathematical domain that studies notions of probability and information via differential geometry. The principle idea is
to find new means of representing information flows, making it possible to describe and reach high-level contents within a real-time audio flow. Easily accessing this information makes diverse artistic and industrial applications possible such as searching musical data, real-time recognition of musical data, and automatic improvisation.

IRCAM's Team: Music Representations team.

  • logo ecole polytechnique
  • logo Ircam
  • logo Thalès RT
  • logo UCSD