Matthieu Kowalski

Demo

Contents

--

Audio Declipping

K. Siedenburg, M. Kowalski and M. Dörfler -- Audio Declipping with Social Sparsity -- ICASSP'14

Software

declipping on GitHub

How to recover the original (blue) signal from the clipped (red) signal ?

Ingredients:

Time-Frequency representations
Sparse Coding
Thresholding rules
Iterative Thresholding

Results (fully unsupervised !):

Double Bass

Guitare

Piano

Audio Source Separation

F. Feng and M. Kowalski -- Revisiting Sparse ICA from a Synthesis Point of View: Blind Source Separation for Over and Underdetermined Mixtures -- Elsevier Signal Processing, Vol. 152, pp 165-177, November 2018

Software

a Matlab Toolbox

How to recover the original sources (speech, instruments...) from the mix ?

Ingredient:

Time-Frequency representations
Sparse Coding
Iterative Thresholding
Indepedence of the sources

Results (fully unsupervised !):

Determined case (Number of sources = Number of mixtures)

Original sources

Mixtures

Estimated sources

Under-determined case (Number of sources > Number of mixtures)

Original sources

Mixtures

Estimated sources

M/EEG Source Localization

A. Gramfort, D. Strohmeier, J. Haueisen, M. Hamalainen and M. Kowalski -- Time-Frequency Mixed-Norms Estimates: Sparse M/EEG imaging with non stationary source activations -- NeuroImage, Vol. 70, pp. 410 - 422, April 2013

Software

More examples with MNE python

How to recover the brain activity from M/EEG measures on the scalp ?

Ingredients:

Time-Frequency representations
(structured) Sparse Coding
Mixed Norms
Iterative Thresholding

Results:

Results obtained with TF- MxNE and MxNE for left-ear auditory stimulation with unfiltered combined MEG/EEG data. The estimation with leads to 2 active space space brain locations at the auditory cortices. TF-MxNE leads to smooth time courses and zeros during baseline.