In this paper, we propose a new extension of LVQ2 based on the Kohonen's feature map algorithm. The neurons of the neural network trained by the Kohonen's feature map algorithm are labeled by using the newly proposed selective learning (SL) algorithm which is the first stage of extended LVQ2. Next, LVQ2 is applied to the neural network labeled by the SL algorithm. And then for further training, the weight vectors of the network are perturbed and finally the LVQ2 algorithm is applied to the network again to complete the extended LVQ2 algorithm. As an application of this extension of LVQ2 algorithm, we construct phoneme classifiers using LVQ2 and extended LVQ2 to compare the performances of the two algorithms. From the phoneme classification tests, we obtain the recognition rates of 60. 4% and 65. 4% for the LVQ2-based and extended LVQ2-based systems, respectively. Keywords: Neural Network, Kohonen's Feature Map, LVQ2, Selective Learning, Extended LVQ2, Phoneme Classification.
Cite as: Kim, I.K., Lee, H.S. (1991) An extended LVQ2 algorithm and its application to phoneme classification. Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991), 1265-1268, doi: 10.21437/Eurospeech.1991-290
@inproceedings{kim91_eurospeech,
author={Il K. Kim and H. S. Lee},
title={{An extended LVQ2 algorithm and its application to phoneme classification}},
year=1991,
booktitle={Proc. 2nd European Conference on Speech Communication and Technology (Eurospeech 1991)},
pages={1265--1268},
doi={10.21437/Eurospeech.1991-290}
}