Robust Speech Recognition of Uncertain or Missing