Error Correction by Paying Attention to Both Acoustic and Confidence References for Automatic Speech Recognition