A multiresolution information measure approach to speech recognition