In this paper, we evaluate feature learning in the problem of retrieving subjectively interesting sounds from electronic music tracks. We describe an active learning system designed to find sounds categorized as samples or loops. These retrieval tasks originate from a broader R&D project, which concerns the use of machine learning for streamlining the creation of videogame content synchronized with soundtracks. The method is expected to function in the context of limited data availability, and as such cannot rely on supervised learning of what constitutes an "interesting sound". We apply an active learning procedure that allows us to find sound samples without predefined classes through user interaction, and evaluate the use of neural network feature extraction in the problem.