Multiresolution information measures applied to speech recognition