Speech signal restoration using an optimal neural network structure

Proceedings of International Conference on Neural Networks (ICNN'96)Published 23 December 2002

Xiaoming Gao, S.J. Ovaska, Z.O. Hartimo

DOI10.1109/icnn.1996.549181

Citations10

TL;DR

An optimal neural network-based method for noisy speech restoration that uses a feedforward neural network with one hidden layer as a nonlinear predictive filter and applies the predictive minimum description length principle to determine the optimal number of input and hidden nodes.

Abstract

In this paper, we propose an optimal neural network-based method for noisy speech restoration. The method uses a feedforward neural network with one hidden layer as a nonlinear predictive filter. In order to select the optimal network structure, we apply the predictive minimum description length principle to determine the optimal number of input and hidden nodes. In this way, the possible over-fitting and under-fitting problem can be penalized automatically. This results in a computationally efficient network structure with both excellent noise attenuation and generalization capabilities.

Keywords

Computer ScienceEngineering

AutomaticaModeling by shortest data description

5,944 Citations1978J. Rissanen

The number of digits it takes to write down an observed sequence x1,...,xN of a time series depends on the model with its parameters that one assumes to have generated the observed data.

The Annals of StatisticsStochastic Complexity and Modeling

945 Citations1986J. Rissanen

View PDF

Bulletin of the London Mathematical SocietySTOCHASTIC COMPLEXITY IN STATISTICAL INQUIRY

651 Citations1991A. P. Dawid

IEEE Transactions on Neural NetworksNetwork information criterion-determining the number of hidden units for an artificial neural network model

649 Citations1994Noboru Murata, S. Yoshizawa +1 more

The problem of model selection, or determination of the number of hidden units, can be approached statistically, by generalizing Akaike's information criterion (AIC) to be applicable to unfaithful models with general loss criteria including regularization terms.

Neural NetworksA new approach for finding the global minimum of error function of neural networks

256 Citations1989Norio Baba

The random optimization method of Matyas and its modified algorithm are used to learn the weights and parameters in a neural network in order to find the global minimum of error function of neural networks.

IEEE Transactions on Neural NetworksAn information criterion for optimal neural network selection

184 Citations1991David B. Fogel

IEEE Transactions on Automatic ControlInconsistency of the AIC rule for estimating the order of autoregressive models

169 Citations1980R.L. Kashyap

The inconsistency of the Akaike information criterion (AIC) rule and its variants for estimating the unknown order of the autoregressive model obeyed by a time series is demonstrated.

Noise reduction using connectionist models

132 Citations2003Shinichi Tamura, Alex Waibel

Using a back propagation network learning algorithm, a four-layered feed-forward network is trained on learning samples to realize a mapping from the set of noisy signals a set of noise-free signals, showing that the network can indeed learn to perform noise reduction.

Robustness against noise: The role of timing-synchrony measurement

37 Citations2005Oded Ghitza

A modified EIH model has been created in which the cochlear filters have been replaced by the uniformly shaped Hamming filters, and the results demonstrate that the timing-synchrony measurement is significantly more robust against noise compared to the power measurement.