Fast learning-algorithms for a self-optimising neural network with an application to isolated word recognition

A short description of the feature finding neural net (FFNN) for the recognition of isolated words is given. As has been shown in the literature, during recognition model FFNN is faster than the classical HMM and DTW recognisers and yields similar recognition rates. In the paper, the emphasis is placed on optimal and fast algorithms for selecting features from the speech signal that are relevant for isolated word recognition. Using the growth algorithm, it is possible to increase the network's size gradually by adding relevant feature detecting cells. The substitution algorithm starts with a full-size net and arbitrary features. Then it replaces less relevant features with features with higher relevance. Recognition results for both cases are given and discussed