ANN-based speech recognition using a preprocessor for non-linear time compression