Project Topics
Get Complete Project Material File(s) Now! »
Table of contents :
Chapter 1 Resume etendu
1.1 Decodage incertain
1.1.1 Etat de l’art
1.1.2 Propagation de l’incertitude comme une matrice de covariance pleine
1.1.3 Estimation de l’incertitude basee sur l’apprentissage generatif
1.1.4 Estimation de l’incertitude basee sur l’apprentissage discriminant
1.2 Apprentissage de reseaux de neurones
1.2.1 Etat de l’art
1.2.2 Approche proposee basee sur une fonction auxiliaire
Chapter 2 Introduction
2.1 Motivation .
2.2 Overview of the problems
2.3 Contributions .
2.4 Outline of the thesis
Chapter 3 Overview of robust ASR
3.1 Overview of speech recognition systems
3.1.1 Pre-processing and feature extraction
3.1.2 Acoustic modeling
3.1.3 Lexicon and language modeling
3.1.4 Decoding .
3.2 Overview of noise-robust ASR
3.2.1 Front-end approaches
3.2.2 Back-end approaches
3.2.3 Hybrid approaches
3.3 Noise-robust ASR datasets
3.3.1 Categorization .
3.3.2 CHiME dataset .
3.4 Summary .
Chapter 4 State of the art
4.1 General framework
4.2 Multichannel source separation
4.3 Uncertainty estimation
4.4 Uncertainty propagation
4.4.1 To the magnitude spectrum
4.4.2 To the static MFCCs
4.4.3 To the dynamic features
4.4.4 Cepstral mean normalization
4.5 Uncertainty decoding
4.5.1 Uncertainty decoding
4.5.2 Modied imputation
4.5.3 Uncertainty training
4.6 Baseline system for CHiME
4.6.1 Speech enhancement baseline
4.6.2 ASR baseline for Track 1
4.6.3 ASR baseline for Track 2
4.7 Upper bound on the ASR performance
4.7.1 Oracle uncertainty
4.7.2 Experimental results
4.7.3 Summary .
Chapter 5 Extension of uncertainty propagation to a full covariance matrix
5.1 Motivation .
5.2 Extension of uncertainty propagation
5.2.1 To the magnitude and the power spectra
5.2.2 To the static MFCCs and the log-energy
5.2.3 To the full feature vector
5.3 Experimental results
5.4 Summary .
Chapter 6 Generative learning based uncertainty estimator
6.1 Motivation .
6.2 Proposed fusion and nonparametric estimation framework
6.2.1 Fused/nonparametric uncertainty estimation
6.2.2 Fused/nonparametric uncertainty propagation with diagonal covariance
6.2.3 Fused/nonparametric uncertainty propagation with full covariance
6.3 Learning of fusion/nonparametric coecients
6.3.1 Weighted divergence measures
6.3.2 Multiplicative update rules
6.4 Experimental evaluation on Track 1
6.4.1 Estimated fusion/nonparametric coecients
6.4.2 ASR results .
6.4.3 Accuracy of uncertainty estimation
6.5 Experimental evaluation on Track 2
6.5.1 Experimental setup
6.6 Summary .
Chapter 7 Discriminative learning based uncertainty estimator
7.1 Motivation .
7.2 Discriminative uncertainty mapping
7.2.1 General approach
7.2.2 Linear mapping .
7.2.3 Nonlinear mapping
7.3 Experimental results
7.3.1 Linear scaling .
7.3.2 Nonlinear transform
7.4 Summary .
Chapter 8 State of the art
8.1 Neural network architectures
8.2 Training algorithms
8.2.1 Training objective
8.2.2 Stochastic gradient descent
8.2.3 Adaptive subgradient method
8.2.4 Back propagation training with second order methods
8.3 Summary .
Chapter 9 Fast neural network training based on an auxiliary function technique
9.1 Motivation .
9.2 Background .
9.2.1 Objective function
9.2.2 Auxiliary function technique
9.3 Quadratic auxiliary function for neural network
9.3.1 First quadratic auxiliary function
9.3.2 Second auxiliary function for separating variables
9.3.3 Recursively deriving auxiliary functions
9.4 Algorithms .
9.4.1 Auxiliary function based NN training
9.4.2 Hybrid algorithm
9.5 Experimental evaluation
9.6 Summary .
Chapter 10 Conclusion and perspectives
10.1 Conclusion .
10.2 Perspectives .
A.1 Comparison of uncertainty decoding on oracle uncertainties on Track 1
A.2 Comparison of uncertainty decoding on the oracle uncertainties on Track 2 .
A.3 Comparison of uncertainty decoding of static and dynamic features on Track
A.4 Comparison of uncertainty decoding of various fusion or nonparametric mapping schemes .
A.5 ASR performance with ROVER fusion
A.6 Comparison of ASR performance on Track 2 of the 2nd CHiME Challenge with GMM-HMM acoustic models
A.7 Comparison of ASR performance on Track 2 of the 2nd CHiME Challenge with a DNN acoustic model
A.8 ASR performance with dicriminative uncertainty estimator
Bibliography
