A simple word-recognition network with the ability to choose its own decision criteria

Various reliable algorithms for the word classification problem have been developed. All these models are necessarily based on the classification of certain `features´ that have to be extracted from the presented word. The general problem in speech recognition is: what kind of features are both word dependent as well as speaker independent? The majority of the existing systems requires a feature selection by the designer, so the system cannot choose the features that best fit the above mentioned criterion. Therefore, the authors tried to build a neural network that is able to rank all the features (here: the cells of the input layer) according to their functional relevance. This method reduces both the necessity to preselect the features as well as the numerical effort by a stepwise removal of the cells that proved to be unimportant