Improvements of the modified hypermap architecture for speech recognition

We are successfully using a coupled modeling concept which is suitable to reproduce sensory systems in a plausible way. The idea was to connect a biological relevant neural network as a filter with a modified hypermap architecture (introduced by Kohonen) to classify the filtered data. The biological filter has the capacity to make segmentation and feature extraction of input data like a sensory pathway. Thus temporal processes can be represented in a sequence of prefiltered segments. The produced time sequences are suited for the analysis by a multistage construction of the modified hypermap architecture. In this paper we present a further improvement of our modified hypermap architecture, the a priori knowledge learning algorithm. The presented system is successfully used for spoken speech recognition of different subjects