Thanks for sharing that! And Pandora just sold for 3.5 billion with ~80 million subscribers, and 6 million paying subscribers!
Pandora’s finger print fell to a lot of the same problems other matching systems do; it’s subjective, and ultimately has to do broad assumptions to avoid false positives. Pandora was great for its time.
Bard, as a product, is primarily an aggregator to get a grip on the ballooning production music industry. Anyone who has had to deal with production knows there are tons of sources for music. We’re trying to do what Kayak did for travel, and create a platform that makes it easier to shop across multiple catalogs. We have about 1.4 million songs that we can query currently.
We have a proprietary data system that we call koreography to fingerprint music. We actually identify things like beat, song structure, instrumentation, and map it in music time. Think of it like MIDI markup, but it can contain more descriptive information.
Our hope is to algorithmically derive the information using the latest MIR (music information retrieval), map it against the timing of the music, and use machine learning to make observation about similarities between music.