Optical coherence tomography (OCT) suffers from the inherent speckle noise in its imaging process, which severely degrades the quality of OCT images. To address such an issue, this paper proposes an asymmetric despeckling generative adversarial network (ADGAN) for OCT speckle noise reduction, based on an unsupervised learning scheme utilizing unpaired clean and noisy images. Specifically, the OCT image despeckling problem is treated as an image-toimage translation problem first, and then the speckle noise reduction is achieved by transferring the noisy images from the noisy domain to the clean domain. Moreover, considering the fact that the information within the clean domain and the noisy domain are imbalanced, an information balancing factor is introduced to capture residual noisy information and help to generate high quality despeckling results. Experimental results show our method surpasses the other state-of-the-art despeckling methods regarding quantitative evaluation metrics and visual qualities.