Synthetic Cross-accent Data Augmentation for Automatic Speech Recognition