Skip to Main content Skip to Navigation
Theses

Décomposition de spectrogrammes musicaux informée par des modèles de synthèse spectrale. Modélisation des variations temporelles dans les éléments sonores.

Romain Hennequin 1
1 AudioSig
TSI - Département Traitement du Signal et des Images, LTCI - Laboratoire Traitement et Communication de l'Information
Abstract : This thesis proposes new methods for automatic musical spectrograms decomposition. Proposed decompositions are derived from Non-negative Matrix Factorization (NMF), which is a powerful rank reduction method renowned for providing decompositions on a few frequency patterns (extracted from the data) that generally have a perceptual meaning. However, NMF cannot efficiently model some kinds of temporal variations of non-stationary events usually found in musical spectrograms. This thesis proposes to introduce generative models of musical spectrograms relying on simple models of sound synthesis in order to take two kinds of common variations into account: the spectral envelope variations (plucked strings sounds, vocal-like sounds...) and the fundamental frequency variations (vibrato, prosody...). Introducing simple synthesis models in factorization methods makes it possible to propose decompositions able to model such variations: a source/filter model permits to take spectral variations of musical objects over time into account. A model of spectrogram with parametric harmonic atoms inspired by additive synthesis and another model inspired by wavetable synthesis which uses transformations of a single atom in order to generate all the possible fundamental frequencies of each instrument make it possible to model fundamental frequency variations. Applications of these new methods are also presented: an application of source separation and one of selective transformation of sound.
Complete list of metadatas

Cited literature [119 references]  Display  Hide  Download

https://pastel.archives-ouvertes.fr/pastel-00648997
Contributor : Romain Hennequin <>
Submitted on : Tuesday, December 6, 2011 - 5:27:08 PM
Last modification on : Friday, July 31, 2020 - 10:44:09 AM
Long-term archiving on: : Wednesday, March 7, 2012 - 3:05:43 AM

Identifiers

  • HAL Id : pastel-00648997, version 1

Collections

Citation

Romain Hennequin. Décomposition de spectrogrammes musicaux informée par des modèles de synthèse spectrale. Modélisation des variations temporelles dans les éléments sonores.. Traitement du signal et de l'image [eess.SP]. Télécom ParisTech, 2011. Français. ⟨pastel-00648997⟩

Share

Metrics

Record views

725

Files downloads

1630