A scaly artificial neural network for speaker independent isolated word recognition using non-linear time alignment

The use of neural networks for the recognition of isolated letters from the English alphabet is investigated. A scaly architecture neural network model is used and is trained using the error backpropagation algorithm. The architecture is varied by changing the number of inputs to the network, the number of hidden units in the network and the performance of each of these networks is compared. The nonlinear time alignment algorithm called trace segmentation is used. The use of different transfer functions within the processing units of the neural network is looked at and several versions of the sigmoid function compared.<
>