Images and sounds from the Sylvain LESAGE PhD Thesis

Figures and sounds form the PhD thesis

"Structured dictionary learning for the sparse modeling of multichannel signals"

Sylvain LESAGE

Chapter 1 - Sparse representation of signals (Représentation parcimonieuse des signaux)

Figure 1.1	eps	pdf
Figure 1.2 left	eps	pdf
Figure 1.2 right	eps	pdf
Figure 1.3	eps	pdf
Figure 1.4	eps	pdf
Figure 1.5	eps	pdf
Figure 1.6	eps	pdf
Figure 1.7	eps	pdf

Violin signal represented on figure 1.4 (full) and on figure 1.1 (100 ms, between 1 s and 1.1 s)

Violin sound

wav

Chapter 2 - Sparse decomposition algorithms (Algorithmes de décomposition parcimonieuse)

Figure 2.1

eps

pdf

Chapter 3 - Dictionary learning (Apprentissage de dictionnaire)

Chapter 4 - Sparse decompositions for source separation (Décompositions parcimonieuses pour la séparation de sources)

Figure 4.1	eps	pdf
Figure 4.2 left	eps	pdf
Figure 4.2 right	eps	pdf

Bass viol signal represented on figure 4.1

Bass viol sound

wav

Mixture of four sources represented on figure 4.2

Mixture	wav
Source 1	wav
Source 2	wav
Source 3	wav
Source 4	wav

Chapter 5 - Critical analysis of the model of non-structured dictionary and proposition (Analyse critique du modèle de dictionnaire non-structuré et proposition)

Figure 5.1	eps	pdf
Figure 5.2	eps	pdf

Chapter 6 - Example of structure : union of orthonormal bases (Exemple de structure : union de bases orthonormales)

Figure 6.1	eps	pdf
Figure 6.2	eps	pdf
Figure 6.3	eps	pdf

Chapter 7 - Fundamental structure : dictionary generated by the linear deformation of motifs (Structure fondamentale : dictionnaire généré par déformation linéaire de motifs)

Figure 7.1	eps	pdf
Figure 7.2	eps	pdf
Figure 7.3	eps	pdf
Figure 7.4	eps	pdf
Figure 7.5	eps	pdf
Figure 7.6	eps	pdf
Figure 7.7	eps	pdf
Figure 7.8	eps	pdf
Figure 7.9	eps	pdf
Figure 7.10	eps	pdf
Figure 7.11	eps	pdf

Chapter 8 - Decomposition algorithms on a structured dictionary (Algorithmes de décomposition sur un dictionnaire structuré)

Figure 8.1	eps	pdf
Figure 8.2	eps	pdf

"The bumblebee fly" signal, used in the experiments of the section 8.7

"The bumblebee fly" sound

wav

Chapter 9 - Structured dictionary learning (Apprentissage de dictionnaire structuré)

Figure 9.1	eps	pdf
Figure 9.2	eps	pdf
Figure 9.3	eps	pdf
Figure 9.4	eps	pdf
Figure 9.5	eps	pdf
Figure 9.6	eps	pdf
Figure 9.7 left	eps	pdf
Figure 9.7 right	eps	pdf
Figure 9.8	eps	pdf
Figure 9.9 left	eps	pdf
Figure 9.9 right	eps	pdf
Figure 9.10	eps	pdf
Figure 9.11 left	eps	pdf
Figure 9.11 right	eps	pdf
Figure 9.12	eps	pdf
Figure 9.13	eps	pdf
Figure 9.14	eps	pdf
Figure 9.15	eps	pdf
Figure 9.16	eps	pdf
Figure 9.17	eps	pdf

Experiments of section 9.5 : learning signals

8 000 Hz signal	wav
11 025 Hz signal	wav

Experiments of section 9.6 : learning and test signals

learning signal 1	wav
learning signal 2	wav
learning signal 3	wav
learning signal 4	wav
learning signal 5	wav
learning signal 6	wav
learning signal 7	wav
learning signal 8	wav
learning signal 9	wav
learning signal 10	wav
test signal	wav

Chapter 10 - Source separation : source estimation using structured dictonaries (Séparation de sources : estimation des sources par dictionnaire structuré)

Figure 10.1	eps	pdf
Figure 10.2 left	eps	pdf
Figure 10.2 right	eps	pdf
Figure 10.3 left	eps	pdf
Figure 10.3 right	eps	pdf
Figure 10.4	eps	pdf
Figure 10.5	eps	pdf
Figure 10.6 left	eps	pdf
Figure 10.6 right	eps	pdf
Figure 10.7 left	eps	pdf
Figure 10.7 right	eps	pdf
Figure 10.8	eps	pdf
Figure 10.9 left	eps	pdf
Figure 10.9 right	eps	pdf
Figure 10.10 left	eps	pdf
Figure 10.10 right	eps	pdf

Experiments in section 10.3 : multichannel separation

LEARNING	Mixture	Source 1	Source 2	Source 3
Original sources	wav	wav	wav	wav
Sources estimated by DUET		wav	wav	wav
Sources estimated by BZ		wav	wav	wav
Sources estimated by DP with Gabor 1		wav	wav	wav
Sources estimated by DP with Gabor 2		wav	wav	wav
Sources estimated by DP avec Learnt		wav	wav	wav

TEST	Mixture	Source 1	Source 2	Source 3
Original sources	wav	wav	wav	wav
Sources estimated by DUET		wav	wav	wav
Sources estimated by BZ		wav	wav	wav
Sources estimated by DP with Gabor 1		wav	wav	wav
Sources estimated by DP with Gabor 2		wav	wav	wav
Sources estimated by DP with Learnt		wav	wav	wav

Experiments in section 10.4 : monochannel separation

learning signal 1 (2 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

learning signal 2 (3 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

learning signal 3 (5 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

learning signal 4 (9 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

learning signal 5 (14 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

learning signal 6 (24 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

learning signal 7 (40 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

learning signal 8 (1 m 07 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

learning signal 9 (1 m 51 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

learning signal 10 (3 m 06 s)	Learning mixture	Learning source 1	Learning source 2	Test mixture	Test source 1	Test source 2
Original sources	wav	wav	wav	wav	wav	wav
Sources estimated by GMM 32		wav	wav		wav	wav
Sources estimated by GMM 64		wav	wav		wav	wav
Sources estimated by DP Learnt		wav	wav		wav	wav

Chapter 11 - Open problems and perspectives (Problèmes ouverts et perspectives)

Figure 11.1

eps

pdf

Chapter 12 - Perpectives for other applications (Perspectives pour d'autres applications)

Figure 12.1

eps

pdf