Pandora is considered the world’s most powerful music discovery platform, using its
Using Data to Teach Computers How to Listen to Music
Music is an art form and behind it are certain objective properties, what instruments went into making it, the overall sound style, and artistic expressions of intent. It’s not just about objectively understanding music itself. We’re using that data to teach computers how to listen to music.
There are about 450 traits that we’re looking at for every song. Is it a breathy voice like Bjork or perhaps a smooth vocal like Shaday? Is there a swing or a shuffle feel to the beat or is there a lot of syncopation? There’s always this talk about AI and machine learning replacing humans when in reality it’s a loop. I think really a lot of the future is those two working together.
Challenge is Determining When to Introduce a New Song
The machine can tell you pretty well that someone’s singing in a piece of music. A machine might have a little harder time telling you what language that person is singing. The type of music we listen to and how we behave when we listen to it can really tell us a lot about ourselves. What do you do the first time you’ve ever heard a song? Are you open to it do you follow the mainstream or do you have very niche interest? Underlying all of those core behaviors are some really fundamental personality traits.
The challenge with music discovery is when is the right moment to hear a new song. That’s where you have to start learning about people. If you’re working and very concentrated you want to be listening to something familiar and hearing something you’ve never heard before is oftentimes a bit jarring.
Biggest Misconception About Music Streaming and Data
The biggest misconception about music streaming and data is that everything behind it is an algorithm. When we fundamentally think about how the algorithms work we have to think back to how we used to discover music before we had streaming services. We would listen to the radio or we would learn from our friends.
What’s at the heart of a lot of these algorithms is actually connecting you to all of the other people out there that you’re never ever going to meet. The trick then becomes how do you combine all of those sources of recommendations with your past listening behavior and with your current circumstance to pick the best song for you right now?