This manuscript describes my main research contributions since 2016 as an Associate Professor ("Maître de Conférences") in the Algomus team (University of Lille, CRIStAL) as well as their perspectives for the coming years. These works are in the field of Music Information Retrieval (MIR) and focus on developing computer-based methods to assist in the analysis and composition of music. The proposed methods are based on machine learning algorithms and focus on symbolic musical representations, which translate the content of the score. The contributions and perspectives of these works are organized along three axes. The first axis focuses on computer-assisted music analysis in the classical music repertoire. It includes works on modeling stylistic elements that are characteristic of this repertoire, around harmony, texture, and structure. The second axis brings together works on modeling guitar tablatures in the modern popular music repertoire. We will present, in particular, a corpus statistical study as well as methods for modeling musical function and proposing tablature continuation by texture imitation, opening perspectives to assist music composition. The third axis presents a set of reflections and experiments aiming at evaluating the contributions of the field of Natural Language Processing (NLP) for modeling musical scores. We will present, in particular, studies regarding the transposition of the mutual attention mechanism in the musical domain and regarding the musical expressiveness of representation systems inspired by NLP.
defended on 27/03/2023
français
English