Nos partenaires

CNRS

Rechercher





Accueil du site > Français > Evénements > Séminaires

Séminaires

 

L’IRIT étant localisé sur plusieurs sites, ses séminaires sont organisés et ont lieu soit à l’Université Toulouse 3 Paul Sabatier (UT3), l’Université Toulouse 1 Capitole (UT1), l’INP-ENSEEIHT ou l’Université Toulouse 2 Jean Jaurès (UT2J).

 

STFT phase recovery based on sinusoidal modeling for audio source separation

Paul MAGRON - Tampere University of Technology (Finlande)

Vendredi 14 Septembre 2018, 16h00 - 17h15
INP-ENSEEIHT, Salle des thèses
Version PDF :

Abstract

For audio source separation applications, it is common to apply a Wiener-like filtering to a time-frequency (TF) representation of the data, such as the short-time Fourier transform (STFT). However, this approach, in which the phase of the original mixture is assigned to each component, is limited when sources overlap in the TF domain. We then address the problem of STFT phase reconstruction for audio source separation.
The first part of this talk will briefly introduce the most common phase recovery techniques, and will highlight the need for novel reconstruction methods. Our approach consists in basing those new methods on phase information that originates from signal models instead of only exploiting transformation-based properties. We present the core-model used in our work, namely the model of mixtures of sinusoids. From this model can be obtained some phase constraints that can further be integrated in mixture models.
In the second part, we will present deterministic approaches for source separation that exploit the afore-mentioned phase model: the magnitudes of the sources are either assumed to be known, or jointly estimated in a complex nonnegative matrix factorization framework.
In the third part, we will exploit those phase models in a probabilistic framework. We design a novel source model in which the phase is no longer assumed uniform, which allows us to account for phase priors. We also combine signal-based and representation-based phase constraints in a unified framework for improved phase recovery. Finally, we present some recent developments on an end-to-end probabilistic model in which both magnitude and phase parameters can be jointly estimated.

Bio: Dr. Paul Magron received the State Engineering degree from the École des Ponts ParisTech, Paris, France, in 2013, and the M.Sc. degree in acoustics, signal processing and computer science applied to music from the University of Paris 6, Paris, France, in 2013. In 2016, he received the Ph.D. degree in audio signal processing from Télécom ParisTech. Since January 2017, he is working as a postdoctoral researcher in the Laboratory of Signal Processing, Tampere University of Technology (Tampere, Finland). His research topics include musical source separation, nonnegative matrix factorization, phase recovery and probabilistic modeling.

 

Retour