Back-propagation training of a neural network for word spotting

An approach to backpropagation training of a neural network for word spotting is described. It is assumed that the network has one output unit for each keyword to be detected, and that features of the speech signal are input at fixed intervals. The goal of training is to obtain a network that emits a detection pulse at the appropriate output unit when the utterance of a keyword is completed. The authors have developed a successful backpropagation strategy which incorporates `don´t care´ targets for outputs expected to be in the process of rising or falling, propagation of errors for only a subset of those times at which no detection pulse is expected, iterative refinement of the temporal placement of target outputs, and use of a super-squared error criterion. In an application of the strategy to speaker-dependent, continuous digit recognition (i.e., digit spotting with no utterances of nondigits), word-error rates of 0% and 2.5% were achieved for the training and test utterances, respectively