Neural network based continuous speech recognition by combining self organizing feature maps and Hidden Markov Modeling

Abstract

This paper describes the investigation of possibilities for using self organizing feature maps in connection with Hidden Markov Modeling (HMM) in order to build a Neural Network based continuous speech recognition system. Starting with a brief outline of the problems arising with the use of Neural Networks for continuous speech recognition and the various attempts in order to solve these problems, the motivation for using self organizing feature maps in combination with Hidden Markov Models is explained. The various aspects and interpretations resulting from that approach are discussed. A description of the details that have to be considered during the design of the feature map and which seem to be of special importance for the use in combination with a Markov model is given. The results obtained with that approach are evaluated and compared to the performance obtained with the use of ordinary HMM based speech recognition algorithms. A final evaluation of the basic idea and conclusions resulting in recommendations for future research directions are given at the end of the paper.