Improving Domain-independent Cloud-Based Speech Recognition with Domain-Dependent Phonetic Post-Processing