🇫🇷 Français
🇬🇧 English
PhD 2024-2027 – Modeling and Semi-Automatic Generation of Musical Arrangements to Foster Ensemble Music Practice
- PhD 2024-2027 in modeling and computer music
- Keywords: computer music, digital humanities, cultural heritage, instrumentation, pedagogy, machine learning
- Location: Lille (CRIStAL, CNRS, UniversitĂ© de Lille, Villeneuve d’Ascq)
- Supervisors: Mathieu Giraud (Directeur de recherche CNRS, CRIStAL, Université de Lille) and Florence Levé (Professeure Univ. Picardie Jules Verne, MIS, Amiens)
- Applications open until April 22, 2024 (CV and cover letter, by email)
The Algomus computer music team, from the CRIStAL laboratory (CNRS UMR 9189, UniversitĂ© de Lille), in collaboration with MIS (UPJV, Amiens), focuses on the analysis and computer generation of music, particularly on its symbolic representations (sheet music, chord charts, tablatures…).
Context
Instrumental pedagogy aims to teach young (and adult) learners instrumental technique, but equally the joy of playing, alone or with others, and thus discovering musical repertoire in its diversity, whether medieval themes, Renaissance, classical and romantic periods, jazz, pop, world music, etc. For this purpose, students in their first years of instrumental study generally play pedagogical arrangements, which are simplified versions of existing music pieces. Arranging is a full-fledged professional practice, but many instrumental teachers have some understanding and create quality arrangements for their students to help them explore different musical repertoires.
Arranging is particularly practiced for ensembles of players, whether in music schools, community settings, or in private, family, or friendly circles. What can a flutist with a few years of experience, a beginner trumpeter, and a skilled amateur guitarist play together? Creating or even just selecting suitable sheet music is often tedious in an amateur setting. Some publishers or websites offer duets, trios, or other accessible combinations, but it is rare to find the desired combinations. Finding such suitable sheet music serves a musical purpose, contributes to cultural heritage, and also serves a social purpose by allowing musicians of different backgrounds to play together.
Objectives
Would it be possible to have an arrangement of a Handel sarabande or a Beatles song for two, three, or four players of different levels? The aim of this thesis is to propose models, algorithms, and a prototype platform to generate such arrangements taking into account the diversity of instruments and levels.
One could certainly consider raw or mixed learning approaches, particularly on conditioned generations. These avenues will be explored, but we will also focus on how procedural generation, coupled with learning, could address this issue. We will aim for high-quality arrangements, created from a meta-arrangement written by a human arranger. What data structures could represent such a meta-arrangement, especially with its textures, melodies, and their variations?
Concretely, the thesis will begin with a state of the art
- in procedural generation,
- in learning and constraint-based generation,
- and notably in conditioning generation methods, by difficulty as well as instrumentation,
- and in texture and in voice separation and identification.
Then the thesis will propose
- models of a flexible, “instrumentable” musical phrase, in interaction with arrangers
- the design, implementation, and evaluation of generative model prototypes coupled with a corpus of meta-arrangements.
This thesis will be in collaboration with arrangers, for example, with analysis, writing, and orchestration classes at the conservatories of Lille and Amiens. The corpora, models, and tools created during this thesis will be freely distributed. The public deliverable will be a prototype platform for educational generation, coupled with the Dezrann platform for musical analysis and sharing, allowing the general public to experiment with arrangements in various music genres.
Profile sought
Master in computer science. Skills in algorithms, data science, and generative models. Musical practice and skills in music theory strongly desired (analysis, harmony, composition, instrumentation, music history).
Schedule, publications, and collaborations
The Algomus team is used to publish at major conferences in the field, including ISMIR. The PhD student will be guided in writing and submitting papers to these conferences, whether on his/her own work or on collaborative work inside or outside the team. The PhD student will regularly participate at conferences and other events in the field, and will be encouraged to collaborate with other scientists and artists. Each PhD student in the Algomus team also undertakes an international research stay of 2-3 months during the course of his or her thesis.
The schedule is individualized for each student, but could be as follows:
- T0 to T0+6 months: bibliography
- T0+6 to T0+16: first models for orchestration analysis, corpus modeling and annotation, first paper submission
- During the second year: 2-months stay in a foreign partner university to consolidate the work and open up to new projects
- T0+18 to T0+26: model validation, co-creative generative models, submissions on the personal work as well on collaborative work
- T0+26 to T0+34: thesis writing, possibly new paper submission
- T0+35 and T0+36: PhD defense, new projects and collaborations
Équipe d’accueil, encadrants, et structuration rĂ©gionale
L’Ă©quipe a dĂ©jĂ une bonne expĂ©rience sur ces sujets: Algomus est un cadre idĂ©al pour mener des recherches innovantes en IA et informatique musicale, fondamentales comme appliquĂ©es. L’Ă©quipe est reconnue dans le domaine (organisatrice de la confĂ©rence nationale JIM 2018, membres de comitĂ©s Ă©ditoriaux…). En 2021, l’Ă©quipe a Ă©tĂ© 3è du concours europĂ©en ‘AI Song Contest’: https://www.algomus.fr/before-you-fly. L’Ă©quipe est en lien avec des artistes, des enseignants et des classes de collèges, et des entreprises (Arobas Music Ă Lille, TikTok/ByteDance Ă Londres). La·le doctorant·e bĂ©nĂ©ficiera de cet environnement.
Les encadrants collaborent depuis plusieurs annĂ©es et l’Ă©quipe a collectivement une cinquantaine de publications sur le domaine dans les dernières annĂ©es. Ils sont attentifs Ă la bonne conduite de la thèse tout comme Ă l’Ă©panouissement et Ă l’Ă©quilibre personnel de chacun.
Le sujet contribue aux questions Ă©tudiĂ©es par le CPER Enhance, notamment sur l’axe “dynamique computationnelle des interactions” (Ă©tude de l’interaction des musicien·nes avec un mĂ©ta-arrangement), mais aussi Ă celles du CPER CornelIA et l’Alliance humAIn (travaux sur l’aspect applicatif de l’IA tout comme sur l’émergence de mĂ©thodes explicables) La thèse s’inscrira dans un rĂ©seau de collaborations tissĂ©es par les encadrants et l’Ă©quipe au cours des dernières annĂ©es, que ce soit des collaborateurs acadĂ©miques en informatique musicale tout comme en musicologie, des professeurs de musique, et des laboratoires de R&D.
References
- J. P. Briot, G. Hadjeres, F. Pachet, Deep Learning Techniques for Music Generation - A Survey, 2019
- D. Cope, The Algorithmic Composer. Madison, 2000
- L. Couturier et al., Annotating Symbolic Texture in Piano Music: a Formal Syntax, SMC 2022
- P. Esling, N. Devis, Creativity in the era of artificial intelligence, JIM 2020
- M. Gover et al., Music Translation: Generating Piano Arrangements in Different Playing Levels Gover, ISMIR 2022
- V. SĂ©bastien et al., Score Analyzer: Automatically Determining Scores Difficulty Level for Instrumental e-Learning, ISMIR 2012