GANs and Poses: An interactive generative music installation controlled by dance moves

Autoren Richard Vogl
Hamid Eghbal-zadeh
Gerhard Widmer
Peter Knees
Editoren
Titel GANs and Poses: An interactive generative music installation controlled by dance moves
Buchtitel Proceedings of ISMIR2018 Interactive Machine-Learning for Music (IML4M) @Exhibition
Typ in Konferenzband
Monat September
Jahr 2018
Seiten http://ismir2018.ircam.fr/pages/events-exhibition.html,
SCCH ID# 18094
Abstract

GANs and Poses is an interactive multi-media installation using human poses to control artificial neural networks generating parts for a music arrangement. To provide incentive for interaction, a game-style UI rewards the audience with audio-visual feedback, which in turn drives the progress through the musical arrangement. This work uses a variety of artificial neural networks to (i) detect persons and their body posture, (ii) classify dance moves, and (iii) generate drum as well as melody patterns.