Thesis

  • L. Feugère (2013)
    Gestural control of singing voice synthesis by rules and musical applications
    Pd. D. thesis supervised by Christophe d’Alessandro, Université Pierre et Marie Curie (UPMC), 26/09/2013
    Related audio-visual files ; 2-pages french abstract

    Show abstract

    Abstract: This thesis deals with the production and control modeling of a synthetic singing voice in the context of making a digital musical instrument. Two instruments are presented: the Cantor Digitalis, focusing on singing vowel control and voice individualization, and the Digitartic, which aims at controlling the articulation of Vowel-Consonant-Vowel syllables. Using an augmented graphic tablet, these instruments allow interactive musical applications with fine temporal control of voice production parameters. The relevance of these musical instruments was established through several public performances of the Chorus Digitalis ensemble. The gestures of the musicians were studied along with the musical tasks required for playing the selected repertoire which was composed of traditional world music (baroque choral, North Indian khayal singing) as well as more contemporary pieces. In particular, an experiment was conducted to analyze the ability to control the fundamental frequency of the Cantor Digitalis. Participants were asked to imitate intervals and melodies according to three tempos using three different modalities (one’s own voice, tablet, and tablet with audio feedback). Results showed that precision was better with the tablet modalities than with one’s own voice, while no significant difference was found between the tablet with and without audio feedback. Both instruments have been unified into one Max/MSP application, which provides an audio-visual and interactive educational tool for understanding voice production.

    Hide abstract

  • L. Feugère (2009)
    Automatic segmentation and real-time voice recognition: the case of beatbox
    Master 2 thesis supervised by Bruno Verbrugghe and Pedro Dias Miguel Cardoso, Voxler, Paris

    Show abstract

    Abstract: The aim of this study is to build a system to be used as an instrument controller by the voice. The reasons for the choice of the voiced onomatopoeias —’Poo’ /pu/, ‘Tsi’ /tsi/, and ‘Ka’ /ka/— are discussed. A corpus of these three sounds was established in musical contexts, over twelve speakers, totalling up to 2850 instances. An onset detection based on energy variation in the Mel bands was implemented in Matlab from an Ircam MAX/MSP patch. The onset detection and the features (LPC, ZCR, High Frequency Content and spectral rolloff for the reduced set) are computed in the same frame. Lastly, we present our choice of classifiers to classify the three vocal sounds. With a thirty-two milliseconds computation size window from the onset, we get 97 % of correctly classified instances among the well detected sounds with Multi-layer perceptron, and a 10-cross validation protocol. We also discuss good results with a smaller computation window so that this system can be used in a realtime context. We are currently implementing the real time system in C++.

    Hide abstract

  • L. Feugère (2008)
    Binaural acoustic clues to perceive space dynamically
    Magistere thesis of Université Paris-Sud supervised by Lorenzo Picinali and Brian F.G. Katz, LIMSI-CNRS, Orsay, FR

    Show abstract

    Abstract: How can a blind person perceive the shape of the space where he is walking? We have tried to partially answer this question by looking at some binaural acoustics cues of a walker moving in a changing shape corridor. In this study, related to the Wayfinding project, we analyzed the similarities between: some parameters computed from the acoustic signal measured at the entrance of the two ear canals; the verbal description of the perceived space; and the physical shape of the corridor. These binaural parameters were compared to the ones from an acoustical model of the corridor which were tested by blind people with binaural rendering in a virtual navigation.

    Hide abstract

  • L. Feugère (2007)
    Automatic labelling of tabla sounds
    Master 1 thesis supervised by Perfecto Herrera, UPF, Music Technology Group, Barcelona

    Show abstract

    Abstract: This study deals with the automatic recognition of sounds of tabla, an Indian percussion played by single or double stroke. Along the choice of the database, its labelling, the choice of the descriptors, a script has been written to get the descriptor value matrix of all the database. This values have been trained and tested by the help of different classifiers from the WEKA software. The global quality of the database/descriptors is demonstrated by a score of 94% correctly classified instances. It seems that the choice of taking into account a human hearing model (Meddhi’s hairs model) and that considering tabla music closed to language is quite relevant. LPC coefficients, which are speech articulator descriptors, reach a good rank among the best descriptors of the classification. While most of the classification has been done in a supervised way, the last part of the work gives some clues about a semi­-unsupervised approach using a Cobweb algorithm recently developed by a MTG pHd student.

    Hide abstract

  • L. Feugère (2006)
    Experimental study of nonlinear vibrations of drums cymbals
    Licence 3 thesis supervised by Cyril Touzé, ENSTA ParisTech, Unité de Mécanique, Palaiseau, FR

    Show abstract

    Résumé : Les vibrations des instruments de percussion non-linéaires (famille regroupant les cymbales) présentent des comportements complexes et des caractéristiques propres aux systèmes non-linéaires en conditions normales de jeu. On peut notamment citer le spectre continu et à large bande en vibrations de grande amplitude et la sensibilité aux conditions initiales. Le comportement de la cymbale en condition normal de jeu est la superposition des vibrations de la cymbale engendrés par une infinité d’excitation forcées mono fréquentielles Cette étude expérimentale conduite sur 5 cymbales d’instrumentiste a pour but de caractériser la transition vers le chaos de la vibration de la cymbale lors d’une excitation forcée mono fréquentielle croissante, et de comparer brièvement les résultats avec le model théorique de vibration de coques minces sphériques « idéales ». Pour cela, j’ai travaillé d’abord en régime linéaire pour l’analyse modale de chaque cymbale, c’est- à-dire à faible amplitude de vibration, puis en passant d’une faible à une forte amplitude d’excitation pour parvenir à faire intervenir les effets non-linéaires de la vibration de la cymbale, qui se traduisent par l’apparition de 2 phases successives succédant la phase périodique de vibration : l’une pseudo périodique, et l’autre chaotique. Ne rentrant pas profondément dans l’aspect théorique, je me suis appliqué à décrire le matériel et protocole expérimental, et les problèmes liés à chaque type d’expérience.

    Hide abstract