Issues with uncertainty decoding for noise robust automatic speech recognition