Where collaborative filtering structurally cannot help you, and what a content-based audio embedding returns instead, with the nearest-neighbour arithmetic worked out.
