Learning Continuous Representation of Audio for Arbitrary Scale Super Resolution