Our paper is out, in which we explore different transfer learning methods, pre-trained models, and dataset characteristics to understand the effect on birdsong classification performance, especially out-of-distribution performance.