The use of fuzzy membership in network training for isolated word recognition

A modification to the use of fuzzy membership in the training of an artificial neural network is presented. The modified membership function can be applied to patterns that have a multi-center data structure in the feature space, and is used in network training for isolated word recognition. The results indicate that the network trained using this fuzzy membership function has a better overall recognition rate than either the network trained by the conventional error backpropagation method or the classifier derived from vector quantization