Thesis of Patrice Thibaud

Computational Analysis of Expressive Timing in Solo Flute Performances

No two performances of the same musical work are ever exactly alike. Even when performers follow the same score, they continuously shape timing, articulation, dynamics, and phrasing to convey their own musical interpretation. These variations are at the heart of musical expression. Understanding and modelling these nuances has become a central topic in Music Information Retrieval and, more broadly, in computational performance analysis. This thesis specifically investigates expressive microtiming, i.e., very small-scale temporal variations that contribute to musical expressivity, as well as the specific challenges posed by the solo flute repertoire. First, we introduce two original annotated corpora: Telemann's Twelve Fantasias for solo flute, representative of the Baroque repertoire, and a corpus dedicated to two landmark works of the early twentieth century, Debussy's Syrinx and Varèse's Density 21.5. These datasets include multi-level annotations describing tonal structure, implicit voices, musical motives, and precise note onset times. Next, we develop several music alignment strategies. On the one hand, we introduce a score-to-audio alignment method based on harmonic models adapted to the specific characteristics of the flute; on the other hand, we compare this approach with different audio representations, including chromagrams and neural network-based multi-pitch estimations. We show that alignment errors exhibit temporal biases specific to the representations used and that a simple post-processing step effectively compensates for these biases, significantly improving alignment accuracy. We also propose a new late-fusion strategy for audio-to-audio alignment, based on the consistency of multiple transfer paths, which improves alignment robustness. Finally, using newly proposed metrics based on inter-onset intervals, we analyse aligned data and reveal interpretive strategies related to metric structure, rhythmic grouping, and implicit polyphony. Performers emphasise interleaved voices and musical patterns through systematic temporal asymmetries. These findings open new perspectives for multimodal performance analysis, notably through the integration of breath capture and extended instrumental techniques.

Jury

- M. Mathieu GIRAUD Directeur de recherche Université de Lille Directeur de thèse, - M. Louis BIGO Professeur des universités Université de Bordeaux Co-directeur de thèse, - M. Yann TEYTAUT Maître de conférences Université de Lille Co-encadrant de thèse, - Mme Clarisse DHAENENS Professeure des universités Université de Lille Examinatrice, - Mme Juliette CHABASSIER Chargée de recherche INRIA Bordeaux, en détachement dans l'entreprise Modartt Examinatrice, - M. Jean-Louis GIAVITTO Directeur de recherche IRCAM Rapporteur, - Mme Elaine CHEW Full professor King's College Rapporteure, - Mme Anja VOLK Full professor Utretch University Examinatrice.

Thesis of the team Algomus defended on 23/11/2026