Multi-modal characterisation of clarinet reeds

Hosting labs: Marie Tahon, Bruno Gazengel, Amélie Gaillard, Emmanuel Brasseur
Hosting labs: Le stage se déroule au sein du Laboratoire d’Acoustique de l’Université du Mans (LAUM) pour la partie expérimentale et au LIUM pour la partie analyses.
Place: Le Mans
Contact: Marie.Tahon(at) et Bruno.Gazengel(at)
Application: Send CV + cover letter to Marie Tahon et Bruno Gazengel, before Decembre 15, 2023


Description : Most single reeds are made from reed, although some are made from composite materials or plastic. The mechanical characteristics of cane mean that reeds are mechanical systems whose properties vary from one reed to another, for an identical family of reeds. Musicians can therefore find reeds of extremely different musical qualities in the same package. In fact, the indicators used by reed makers (reed type, strength) are insufficient to characterise a reed musically.

The aim of the reeds project is to determine the characteristic elements that would explain why a musician perceives a reed as being bad or good. This question has already been studied from several angles [1,2], using the mechanical characteristics of a reed, descriptors extracted from audio signals recorded by musicians, or musicians’ perceptions. The effect of asymmetry has also been studied [3].

Objective: The aim of the internship is to complement Amélie Gaillard’s PhD work on this topic in two main ways:

1) Creation of a database consisting of mechanical characteristics measured on several reeds (bin or concert) using an artificial mouthpiece and measurements of static stiffness profiles.

2) Recordings of several musicians (around ten) on several reeds. Several recordings can be made: audio recording in a controlled environment when the musician is playing a given reed, video capture of the face in a playing situation.

3) Relevant descriptors should be extracted from these data: mechanical (stiffness, pressure, threshold pressure, etc.), audio (harmonic descriptors, CGS, buzziness), HNR, etc., video (facial units, distance between these characteristic points, estimated deformation of the mouth), etc.

4) Analysis of the data using the statistical approaches proposed in the article (Petiot, 2017), such as ANOVA, multi-variate analysis, etc. Then another approach based on machine learning will be considered: prediction of reed quality conditioned by the various characteristics recorded. Algorithms such as Gaussian mixture models, support vector machines (SVM) or neural networks will be used.



  • [1] Bruno Gazengel, Jean-Pierre Dalmont, and Jean Francois Petiot. “Link between objective and subjective characterizations of Bb clarinet reeds.” Applied Acoustics 106 (2016): 155-166.
  • [2] Jean-François Petiot, Pierric Kersaudy, Gary Scavone, Stephen Mcadams, Bruno Gazengel. Investigation of the Relationships Between Perceived Qualities and Sound Parameters of Saxophone Reeds. Acta Acustica united with Acustica, 2017, 103 (5), pp.812 – 829.
  • [3] Amélie Gaillard, Vincent Koehl, Bruno Gazengel. Link between stiffness symmetry and perceived quality of clarinet cane reeds. Forum Acusticum 2023, the 10th European Congress on Acoustics, European Acoustics Association, Sep 2023, Turin, Italy.