Boosting Cross-Domain Speech Recognition With Self-Supervision