A structural segmentation of songs using generalized likelihood ratio under regularity assumptions

Gabriel Sargent 1 Frédéric Bimbot 1 Emmanuel Vincent 1
1 METISS - Speech and sound data modeling and processing
IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires, Inria Rennes – Bretagne Atlantique
Abstract : This document presents the algorithm submitted to the "Structural segmentation" task at MIREX 2010. It consists in three parts. First, feature extraction (beat, MFCC, Chroma) from the song is achieved using existing scripts. Second, a segmentation is done according to three criteria for localizing statistical breakpoints, repeated feature sequences, and short events, using a filtered version of generalized likelihood ratio. The segment borders are then selected according to the amplitude of these criteria and a regularity constraint about the length of the structural segments searched. Third, the segments are gathered into similar classes using a hierarchical (agglomerative) clustering. The number of steps of this clustering is estimated separately for each song.
Type de document :
Poster
MIREX - ISMIR 2010, Aug 2010, Utrecht, Netherlands. 2010
Liste complète des métadonnées

https://hal.inria.fr/inria-00551411
Contributeur : Gabriel Sargent <>
Soumis le : lundi 3 janvier 2011 - 16:11:29
Dernière modification le : mercredi 16 mai 2018 - 11:23:03
Document(s) archivé(s) le : lundi 5 novembre 2012 - 15:15:15

Identifiants

  • HAL Id : inria-00551411, version 1

Citation

Gabriel Sargent, Frédéric Bimbot, Emmanuel Vincent. A structural segmentation of songs using generalized likelihood ratio under regularity assumptions. MIREX - ISMIR 2010, Aug 2010, Utrecht, Netherlands. 2010. 〈inria-00551411〉

Partager

Métriques

Consultations de la notice

357

Téléchargements de fichiers

170