de
en
Schliessen
Detailsuche
Bibliotheken
Projekt
Impressum
Datenschutz
Schliessen
Publizieren
Besondere Sammlungen
Digitalisierungsservice
Hilfe
Impressum
Datenschutz
zum Inhalt
Detailsuche
Schnellsuche:
OK
Ergebnisliste
Titel
Titel
Inhalt
Inhalt
Seite
Seite
Im Werk suchen
Integration of neural networks and probabilistic spatial models for acoustic blind source separation / von M. Sc. Lukas Drude ; Erster Gutachter: Prof. Dr.-Ing. Reinhold Häb-Umbach, Zweiter Gutachter: Prof. Dr.-Ing. Timo Gerkmann. Paderborn, 2020
Inhalt
Abstract
Acknowledgments
Introduction
Prerequisites
Notation
Signal model
Overview table of variable names
Random variables
Latent variable models and the expectation maximization algorithm
Latent variable models
Mixture models
Expectation maximization algorithm
Blind source separation principles
Principles of single-channel approaches
Shallow methods
Deep-learning methods
DC: Deep clustering
DAN: Deep attractor network
PIT: Permutation invariant training
Discussion of single-channel deep-learning methods
Principles of multi-channel approaches
Probabilistic spatial mixture models
Frequency permutation problem
Initialization
Influence of the mixture weight
Complex Watson mixture model
Complex Bingham mixture model
Full-Bayesian complex Watson mixture model
Time-variant complex Gaussian mixture model
Complex angular central Gaussian mixture model
Guided source separation
Spatial features for neural networks
Principles of source extraction
Spectral subtraction/ masking
Spatial filtering/ beamforming
Spatial covariance matrix estimation
MaxSNR/GEV
MVDR
Linearly constrained minimum variance beamformer
Weighted multi-channel Wiener filter
Magnitude and phase normalization of beamforming vectors
Combination of beamforming and masking
Integration of neural networks and probabilistic graphical models
Existing integration approaches
Cascade approach: Integration by initialization
Tight integration of spatial and spectral features
vMFcACGMM
Additional constraints
Unsupervised training using multi-channel features
Evaluation
Performance metrics
Database design
WSJ0-2mix
WSJ-BSS
WSJ-MC
Acoustic model training
Deep-learning methods
Deep clustering
Deep attractor network
Permutation invariant training
Comparison with reference publications on WSJ0-2mix
Probabilistic spatial mixture models
Type of the spatial observation model
Parameter choice for the cACGMM
Source extraction
Integration of neural networks and probabilistic graphical models
Weak integration: A cascade approach
Strong integration
Comparison of integration models with single-/ multi-channel encoder
Unsupervised training of deep clustering
Overview of all methods on WSJ-BSS
Analysis of splits of the WSJ-BSS database
Analysis with matched training of the acoustic model
Overview of all methods on WSJ-MC
Reproducibility and statistical significance
Conclusion
Appendix
Properties of the complex Bingham distribution
Eigenvalue shift in the normalization term
Eigenvalue shift in the distribution
Non-negativity of the Kullback-Leibler divergence
Mixture weights without Lagrange's method
Remarks on complex derivatives
GEV/MaxSNR beamformer
Solution with constraint optimization
Solution without constraint optimization
MVDR beamformer
Permutation formalism
Comparison of WSJ-BSS and SMS-WSJ
More detailed evaluation results
Glossary
List of peer-reviewed publications with own contributions (OC)
Bibliography
Die detaillierte Suchanfrage erfordert aktiviertes Javascript.