Learning from Imperfect Data LID Challenge Track2