This folder contains the different works created around Ken Déguernel's Ph.D. Thesis "Learning of Musical Structures in the Context of Improvisation" and DYCI2's second Work Package "Interactive Learning of Musical Structures".
This repository is divided in three parts:
- The multi-dim folder defines methods using a combination of probabilistic models and factor oracle to generate multidimensional improvisations. A journal paper about the methods can be found here.
- The reinforcement folder include works by intern Rémi Decelle on reinforcement learning and the detection of "musical error" (ie. wrong note) during generation. The internship report can be found here.
- The multi-level folder defines methods to automatically analyse the hierarchical structure of a scenario using the notion of mutual information. This method is detailed in Déguernel's Ph.D. thesis.
Author: Ken Déguernel (Inria Nancy Grand Est / IRCAM STMS LAB);
Contributors: Rémi Decelle, Nathan Libermann (Inria Nancy - Grand Est / Inria Rennes); DYCI2 supervision : Gérard Assayag (Ircam, PI), Emmanuel Vincent (Inria Nancy, WP leader).
Licence: GPL v3.