Nicolas Guiomard-Kagan received a Master's Degree in Computer Science and Applications from the University of Rouen. He did his internship of Master 2 on the rational languages k-blocks deterministes at LITIS (Haute-Normandie). Since 2013, Nicolas Guiomard-Kagan has been preparing his thesis at MIS (UPJV, Amiens), and is working on simplifying the analysis of polyphonic partitions by decomposing them into monophonic voices or streams (coherent sets of notes). He is supervised by Mathieu Giraud, Richard Groult and Florence Levé in the Algomus research group.
Scientific Activities
I'm in SDMA team of
MIS laboratory. I'm also
in Algomus research group
between Amiens and Lille University on Music informatics. My area of
research are Music information retrieval and text algorithms.
I work with symbolic data (score) on which I apply techniques derived
from text algorithms.
Music can be either monophonic (a single note sounds at each time) or polyphonic
(several notes sound simultaneously, building harmonies). Understanding polyphonic
music can be very complex. The goal of my research is to ease the analysis of polyphonic
scores by splitting them in either monophonic voices or streams (coherent sets of notes).
I proposed a comparaison between three voices separation algorithms and three streams
separation algorithms. I proposed also an evaluation method to fairly compare these two approaches.
This work was published in the International Society for Music
Information Retrieval Conference (ISMIR 2015).
The conclusions of this study led me to propose an improvement of a voice separation algorithm,
published in proceedings of ISMIR 2016.
One of my goals is now to contribute to the computational music analysis of polyphonic music by combining
some features I have been working on with new machine learning approaches.
Thesis
Abstract
Music can be either monophonic (a single note sounds at each time) or polyphonic (several notes sound simultaneously, building harmonies). Understanding polyphonic music can be very complex. The goal of this thesis in computer music is to ease the analysis of polyphonic scores by splitting them in either monophonic voices or streams (coherent sets of notes). Research in this thesis first consists in comparing three voices separation algorithms and three streams separation algorithms. I propose an evaluation method to fairly compare these two approaches. This study shows the qualities of the Chew and Wu algorithm. The first step of this algorithm, which segments the score into “contigs” having a constant number of voices, is particularly robust. Further work of this thesis focuses on the second stage of the Chew and Wu algorithm that defines what contigs to connect and how to connect them. I improve these connections by using musical parameters such as the average pitch di erence between neighbor contigs. The thesis concludes by evaluating simultaneously voice separation and pattern matching for the music analysis of fugues. Links to the thesis in French and the slides in English.Publications
Article of international conference with reviewing committee
-
N. Guiomard-Kagan, M. Giraud, R. Groult, F. Levé, Comparing Voice and Stream
Segmentation Algorithms, In International Society for Music Information Retrieval
Conference (ISMIR 2015) (pp. 493-499). (acceptance rate : 47.1%, 114 accepted papers
on 242 reviewed).
This paper, published in a major music informatics conference, presents a comparative work of 5 partition segmentation algorithms. I proposed a way to simultaneously compare two algorithms of voice separations and three algorithms of separations in streams. -
N. Guiomard-Kagan, M. Giraud, R. Groult, F. Levé, Improving Voice Separation by
Better Connecting Contigs, In International Society for Music Information Retrieval
Conference (ISMIR 2016) (pp. 164-170). (acceptance rate : 47.5%, 113 accepted papers
on 238 reviewed).
One of the best algorithm for the problem of voices separation is Chew and Wu 2005, where first the partition is segmented in contigs (in each contig, the number of voices is constant), then the contigs are connected to find the voices. In this article, we have improved the order of connection between the contigs, and also the way to connect the contigs. The graphs in the figure below show the percentage of merging errors performed in relation to the total number of connections, it can be seen that using our parameters (green curves) we make fewer errors.
Article of national conference with reviewing committee
|
Collaboration
- February 2016, exchange with the University of Thessaloniki (Greece). I went to Thessaloniki for one week to work in collaboration with the Cognitive and Computational Musicology Group of the University of Thessaloniki, headed by Emilios Cambouropoulos. With Dimos Makris (Phd student), we were interested in the problem of segmentation in streams.
Software development
- Participate at the development of a tool of music21 (python library helping musical analysis). My participation was part on comparisons methods.
Scientific diffusion
Communications / talk
- October 2016, young scientists research school (EJCIM) in Computer science (GdR IM) --> talk
- December 2014, musical analysis french conference (JAM) --> poster
- June 2016, organisation of young scientists day of MIS laboratory
- Juin 2016, young scientists day of MIS --> talk
- October 2015, young scientists day of MIS --> talk
- June 2014, Seminar of the Combinatorial Team and Algorithms of the LITIS (university of Rouen) --> seminar
- June 2013, young scientists day of MIS --> poster
Popular science, scientific mediation
|
- Mars 2012 : Presentation of my master's thesis on the conjecture of Černý (automates synchronisation) to high school students.
Teaching activities
During the year 2014/2015, I had an extended doctoral contract that allowed me to give 64 hours of teaching. In 2016/2017, I had a temporary teaching and research assistant contract for which I gave 176 hours of teaching. In addition to these two contracts, I gave courses in 2015 and 2016 in an engineering school and in a communication school. Lessons carried (totaling 266 hours) allowed me to discover several facets of the teaching profession, both in the dissemination of knowledge (theoretical course / pratical course) and in the evaluation of knowledge (writing and correction of exams, project defenses, pratical evaluation). Moreover, the courses achieved in school of engineer and school of communication allowed me to teach to different audiences. The tutoring I did during my Master 2 allowed me to help a student in difficulty.- 2016-2017: Network, Licence/Bachelor 2, 18h tutorial, 14h practical course, group size: 20 students.
- 2016-2017 : Internet and programming, Licence/Bachelor 1, 52h tutorial, group size: 35 students.
- 2014-2015 and 2016-2017: Introduction to computer science, Licence/Bachelor 1, 64h tutorial, 42h practical course, group size: 35 students. I had to present the basics of algorithm tools (variables, conditions, loops ...). I also had to help with the preparation and correction of the exams.
- 2014-2015 and 2016-2017: Introduction to databases, Licence/Bachelor 1, 32h tutorial, 12h practical course, group size: 35 students. In this course, the student learned databases principle. In a first time, we presented the entity/association model, after that we continued on the relationnel model. Also we have introduced SQL language.
- 2014-2015: Basic structures, Licence/Bachelor 2, 8h practical course, group size: 20 students. This course consists in making a project in C, the goal was to develop a structure allowing to realise several operations on large numbers. I participated to the projects defenses.
- 2016-2017: Algorithms and Programming, second year of Engineering school, 32h theoretical course, group size: 25 students.
- 2015-2016: Operating systems, fourth year of Engineering school, 12h practical course, group size: 25 students.
- 2015-2016: Multitasking, fourth year of Engineering school, 20h practical course, group size: 25 students. The purpose was to understand the functional organization of an application in cooperating parallel tasks. Students had to develop a structured application in parallel tasks using mechanisms of cooperation (synchronization, communication, mutual exclusion).
- 2015-2016: Web Software, Level License/Bachelor 1 (first year of communication school), 14h tutorial, group size: 15 students. The aim of this course was to give students a broad knowledge base on how to create, build websites, manage newsletter campaigns. I evaluated the students on projects.
- 2015-2016: Deepening of web software, Level License/Bachelor 2 (second year of communication school), 14h tutorial, group size: 15 students.
- 2012-2013: Logic, Master 1, tutoring 8h, group size: 1 student.
- 2012-2013: Compilation, Master 1, tutoring 8h, group size: 1 student.
Administrative Responsibilities
- 2014 - 2016: Elected representative of PhD students in the MIS laboratory of University of Picardie
- Mars 2012 – July 2013: Elected student at council for studies and university life of the university of Rouen.
- February 2013 – July 2013: Elected student at science and technical management board of the University of Rouen