Year: 2015
Type: dissertation
Abstract: This thesis focuses on noise robust automatic speech recognition (ASR). It includes twoparts. First, we focus on better handling of uncertainty to improve the performance of ASR ina noisy environment. Second, we present a method to accelerate the training process of a neuralnetwork using an auxiliary function technique.In the first part, multichannel speech enhancement is applied to input noisy speech. Theposterior distribution of the underlying clean speech is then estimated, as represented by its meanand its covariance matrix or uncertainty. We show how to propagate the diagonal uncertaintycovariance matrix in the spectral domain through the feature computation stage to obtain thefull uncertainty covariance matrix in the feature domain. Uncertainty decoding exploits thisposterior distribution to dynamically modify the acoustic model parameters in the decodingrule. The uncertainty decoding rule simply consists of adding the uncertainty covariance matrixof the enhanced features to the variance of each Gaussian component.We then propose two uncertainty estimators based on fusion to nonparametric estimation,respectively. To build a new estimator, we consider a linear combination of existing uncertaintyestimators or kernel functions. The combination weights are generatively estimated by mini-mizing some divergence with respect to the oracle uncertainty. The divergence measures usedare weighted versions of Kullback-Leibler (KL), Itakura-Saito (IS), and Euclidean (EU) diver-gences. Due to the inherent nonnegativity of uncertainty, this estimation problem can be seenas an instance of weighted nonnegative matrix factorization (NMF).In addition, we propose two discriminative uncertainty estimators based on linear or nonlin-ear mapping of the generatively estimated uncertainty. This mapping is trained so as to maxi-mize the boosted maximum mutual information (bMMI) criterion. We compute the derivativeof this criterion using the chain rule and optimize it using stochastic gradient descent.In the second part, we introduce a new learning rule for neural networks that is based onan auxiliary function technique without parameter tuning. Instead of minimizing the objectivefunction, this technique consists of minimizing a quadratic auxiliary function which is recursivelyintroduced layer by layer and which has a closed-form optimum. Based on the properties of thisauxiliary function, the monotonic decrease of the new learning rule is guaranteed.
Author: Dung Tien Tran
Language: French
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2461132270