A study on integrating acoustic-phonetic information into lattice rescoring for automatic speech recognition