The middle section of Queen’s Bohemian Rhapsody is segmented by onsets, and a feature vector is extracted (via librosa) from each chunk. The samples are then embedded into 2-d using t-SNE. Software used can be found here.
Click to play.