Data Becomes Melody in the Prop Explorer Sound Update - live on Steam! How can data sing? The (very) experimental system uses two deterministic random-projection passes and a K-Means clustering pass to turn each row of data into a stable 16-step musical fingerprint (one bar of melody). First Random Projection is for Pitch:
Each row’s normalized feature values are projected into 16 pitch scores. These are bucketed into harmonic frequencies: C, E, G, or high C. Second Random Projection is for Rhythm:
A separate 16-step projection decides which notes are heard. The four strongest rhythm scores become the audible steps for that prop. K-Means Clustering is for Octave:
Finally, props are grouped into octaves using 5-group K-Means clustering. Rows are clustered by similarity, then each cluster’s centroid average determines whether that group is assigned to a lower or higher octave. The result is a spatial data ensemble where each row gets its own musical voice. Similar rows tend to have similar melodies and voices.