I think a few people have open-sourced various audio fingerprinting algorithms. Perhaps we should standardize around one and enable real UUID for audio files? That seems to be the missing piece for a lot of the novel functionality you're looking to hit. Otherwise you're stuck doing fuzzy pattern matching on metadata, which seems tricky especially considering possible variation in live recording etc.