Prototypical Contrastive Learning For Improved Few-Shot Audio Classification