Roland Badeau: Probabilistic modelling of music time-frequency representations

6 March 2013

Time: 2:00 - 3:00pm
Venue: Eng. 2.09 Engineering Building, Queen Mary University of London, Mile End Road, London, E1 4NS

Nonnegative Matrix Factorization (NMF) is a powerful tool for decomposing mixtures of non-stationary signals in the Time-Frequency (TF) domain. In the literature, a variety of probabilistic models involving latent variables have been designed for introducing some a priori knowledge (like harmonicity and smoothness) into NMF. However, phases are generally ignored in such models, which results in a limited spectral resolution (sinusoids in the same frequency band cannot be properly separated). Moreover, most of these models assume that all TF coefficients are independent, which is not the case of sinusoidal signals for instance. In this talk, I will present a unified probabilistic model called HR-NMF, which achieves a high spectral resolution by taking both phases and local correlations in each frequency band into account. The potential of this new approach will be illustrated in the context of audio source separation and audio inpainting.

Abstract:
Nonnegative Matrix Factorization (NMF) is a powerful tool for decomposing mixtures of non-stationary signals in the Time-Frequency (TF) domain. In the literature, a variety of probabilistic models involving latent variables have been designed for introducing some a priori knowledge (like harmonicity and smoothness) into NMF. However, phases are generally ignored in such models, which results in a limited spectral resolution (sinusoids in the same frequency band cannot be properly separated). Moreover, most of these models assume that all TF coefficients are independent, which is not the case of sinusoidal signals for instance. In this talk, I will present a unified probabilistic model called HR-NMF, which achieves a high spectral resolution by taking both phases and local correlations in each frequency band into account. The potential of this new approach will be illustrated in the context of audio source separation and audio inpainting.

Bio:
Dr. Roland Badeau works as an Associate Professor in the Signal and Image Processing Department, Télécom ParisTech / CNRS LTCI, France, and from February 4 to August 2, 2013, he is visiting the Centre for Digital Music, Queen Mary, University of London. He received the Ph.D. degree from Telecom ParisTech in 2005, and the Habilitation degree from the Université Pierre et Marie Curie (UPMC) in 2010. His research interests focus on statistical modelling of non-stationary signals (including adaptive high resolution spectral analysis and Bayesian extensions to NMF), with applications to audio and music (source separation, multipitch estimation, automatic music transcription, audio coding, audio inpainting). He is a co-author of 19 journal papers, 50 international conference papers, and 2 patents. He teaches in the Master of Engineering of Télécom ParisTech and in the Master of Sciences and Technologies of UPMC. He is also a Senior Member of the IEEE, and an Associate Editor of the EURASIP Journal on Audio, Speech, and Music Processing.