[2211.02712] Resource-Efficient Transfer Learning From Speech Foundation Model Using Hierarchical Feature Fusion