SSHR: Leveraging Self-supervised Hierarchical Representations for Multilingual Automatic Speech Recognition