Modèles granulaires pour les signaux sonores : contributions théoriques et expérimentales

Lorcan Mc Donagh 1
1 METISS - Speech and sound data modeling and processing
IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires, Inria Rennes – Bretagne Atlantique
Abstract : Current sound synthesis techniques are capable of reproducing a great variety of sounds from a small set of variable control parameters. On the other hand, analysis of audio signals is commonly achieved by decomposing the former as a linear combination of a great many objects, typically of sinusoid-like shape, plus optional transient and noise components. Our concern is in replacing these sinusod-shaped objetcs with objects computed using a mathematical function that implements a sound synthesis method. This thesis is devoted to the study of a so-called "granular" model for audio signals. The general characteristics of this model are introduced and discussed in detail, and an example of a practical method for the computation of the model is given. In the rst class of models, which draws upon WaveTable synthesis methods, objects are rst selected amongst a dictionnary of complex wave-shapes, then distorted. The second class of models, referred to as TS, rely on a dictionnary of reference Power Spectral Densities pro les coupled to a parametrised distortion of the phase spectrum. The present work provides a contribution to the elds of signal analysis and classi cation techniques, as well as an original solution to the ecient signal-representation problem, of which audio compression is a direct application. Technical implementation details and the results of an evaluation conducted on real-world audio signals are provided to facilitate further evaluation of the model. This work investigates the possibilities of including elements of musical synthesis into audio signal modelling, and our hope is that it sparks new developments in both domains. 2
Keywords : Signal proc
Document type :
Complete list of metadatas

Cited literature [98 references]  Display  Hide  Download
Contributor : Mc Donagh Lorcan <>
Submitted on : Friday, September 9, 2016 - 3:52:05 PM
Last modification on : Friday, November 16, 2018 - 1:22:51 AM
Long-term archiving on: Saturday, December 10, 2016 - 1:28:41 PM


  • HAL Id : tel-01363446, version 1


Lorcan Mc Donagh. Modèles granulaires pour les signaux sonores : contributions théoriques et expérimentales. Traitement du signal et de l'image [eess.SP]. Rennes 1, 2005. Français. ⟨NNT : 2005REN1S047⟩. ⟨tel-01363446⟩



Record views


Files downloads