Caractérisation de l'environnement musical dans les documents audiovisuels

Abstract : Currently, the amount of music available, notably via the Internet, is growing daily. The collections are too huge for a user to navigate into without help from a computer. Our work takes place in the general context of music indexation. In order to detail the context of our work, we present a brief overview of the work currently made in music indexation for indexation : instrument recognition, tonality and tempo estimation, genre and mood classification, singer identification, melody, score, chord and lyrics transcription. For each of these subjects, we insist on the definition of the problem and of technical terms, and on the more imporants problems encountered. In a second part, we present au method we developped to automatically distinguish between monophonic and polyphonic sounds. For this task, we developped two new parameters, based on the analysis of a confidence indicator. The modeling of these parameters is made with Weibull bivariate distributions. We studied the problem of the estimation of the parameters of this distribution, and suggested an original method derived from the moment method. A full set of experiment allow us to compare our system with classical method, and to validate each step of our approach. In the third part, we present a singing voice detector, in monophonic and polyphonic context. This method is base on the detection of vibrato. This parameter is derived from the analysis of the fundamental frequency, so it is a priori defined for monophonic sounds. Using two segmentations, we extend this concept to polyphonic sound, and present a new parameter : the extended vibrato. Our system's performances are comparable with those of state-of-the-art methods. Using the monophonic / polyphonic distinction as a pre-processing allow us to adapt our singing voice detector to each context. This leads to an improvment of the results. After giving some reflexions on the use of music for automatic description, annotating and indexing of audiovisual documents, we present the contribution of each tool we presented to music indexation, and to audiovisual documents indexation using music, and finally give some perspectives.
Document type :
Complete list of metadatas
Contributor : Hélène Lachambre <>
Submitted on : Wednesday, February 17, 2010 - 3:26:57 PM
Last modification on : Monday, April 29, 2019 - 4:38:23 PM
Long-term archiving on : Thursday, October 18, 2012 - 3:20:10 PM


  • HAL Id : tel-00457522, version 1


Hélène Lachambre. Caractérisation de l'environnement musical dans les documents audiovisuels. Informatique [cs]. Université Paul Sabatier - Toulouse III, 2009. Français. ⟨tel-00457522⟩



Record views


Files downloads