Enhancing Unsupervised Audio Representation Learning via Adversarial Sample Generation