Text this: Robust speech recognition of uncertain or missing data /