An Iterative Framework for Self-supervised Deep Speaker Representation Learning