Improving Accented Speech Recognition with Multi-Domain Training