RECURRENT NEURAL NETWORKS AS PHONEME SPOTTERS

Modules of Recurrent Neural Networks can be used as basic building blocks in models of human word recognition systems, if they exhibit spotting behavior. It is shown that such modules are forced to extract the relevant information from the signal and store it in a time-shift-independent way, if they are trained with a so-called “brute-force-target-function” and if the training data offered to the network is not time-aligned.