Isolated word recognition system using MLP neural network constructive training algorithm

In this paper, a new constructive training algorithm for feed forward MLP neural networks has been developed for isolated word recognition. An incremental training procedure has been employed where the training patterns are learned incrementally, i.e one by one. This algorithm started with a single training pattern and a single hidden-layer using one neuron. During neural network training, the hidden neuron has been increased when the mean square error (MSE) of the training data (TD) is not reduce or the algorithm gets stuck in a local minimum. Input patterns are trained incrementally (one by one) until all patterns of TD are selected and trained.