Publication Details
Parallel training of neural networks for speech recognition
Artificial neural networks, speech recognition, parallel training algorithms
In speech recognition, forward multi-layer neural networks are used as classifiers for phoneme recognizers, for speechparameterization, in language models, and for language or speaker recognition. This paper discusses possibilities oftraining forward multi-layer neural networks using parallel algorithms. The need for parallel training of neuralnetworks is caused by huge quantity of training data used in speech recognition. Synchronous and asynchronousvariants of the training are discussed and experimental results are reported on a real speech processing task.
In speech recognition, forward multi-layer neural networks are used as classifiers for phoneme recognizers, for speech parameterization, in language models, and for language or speaker recognition. This paper discusses possibilities of training forward multi-layer neural networks using parallel algorithms. The need for parallel training of neural networks is caused by huge quantity of training data used in speech recognition. Synchronous and asynchronous variants of the training are discussed and experimental results are reported on a real speech processing task.
@inproceedings{BUT22372,
author="Stanislav {Kontár}",
title="Parallel training of neural networks for speech recognition",
booktitle="Proc. 12th International Conference on Soft Computing MENDEL'06",
year="2006",
pages="6037--6042",
publisher="Brno University of Technology",
address="Brno",
isbn="80-214-3195-4",
url="http://www.fit.vutbr.cz/~cernocky/publi/2006/mendel_2006.pdf"
}