Sort of. They were early to use vector databases for semantic search. Now we use embeddings from a neural net as features for semantic search, I think they used lots of labels and analyzed features like beats per minute, chord progression, etc. happy to be corrected of course.