A Mostly Data-Driven Approach to Inverse Text Normalization

doi 10.21437/interspeech.2017-1274