Attributs multiples dans un improvisateur automatique

TitleAttributs multiples dans un improvisateur automatique
Publication TypeMaster Thesis
Year of Publication2004
AuthorsLaurier, C.
preprint/postprint documentfiles/publications/1e15f3-DEA-2004-Laurier.pdf
AbstractThis master thesis presents the results from my work in the Musical Representations team at Ircam. This research is about modeling the musical style and especially in the context of the improvisation project named "Omax". This study is about a statistical model, the Multi-Attribute Prediction Suffix Trees for style learning and improvisation. This model is built from symbolic sequences from which it should be able to generate a continuation with the same statistical characteristics. Several methods like Hidden Markov Models can produce this kind of result, but the interest of this model is its high level of analysis and management of different attributes. The drawback of this promising method is a high computational cost making it not usable for real-time purposes. In this document we detail this model, the attributes, how to use them and our applications for improvisation. We also present several functions implementing this model in LISP for the Ircam OpenMusic software.
intranet