Music is ubiquitous; it is one of the forces that binds the world together and brings people of all cultures under one umbrella. The novelist Maya Angelou once said, “Everything in the universe has a rhythm, everything dances”. Whether it be through live concerts, street melodies, the radio, or with a pair of headphones plugged into an iPod or computer, people listen to music all across the globe.
Pandora radio, an internet service that plays tunes based on the listener’s favorite songs (Pandora jargon labels these as “radio stations”), has over 23 million subscribers. Incidentally, I am currently on Pandora as I write this blog post, listening to songs from one of my sixteen stations. The unique part of Pandora is that it selects songs that are similar in rhythm, tune, harmonies, beats–basically Pandora is able to dissect a song into its musical components, categorize these components, and then choose another song that is matches these categories.
The science behind Pandora lies in the Music Genome Project, which is an undertaking that began thirteen years ago, when Tim Westergren, the head of Pandora Radio, teamed up with Nolan Gasser, who is currently the Chief Musicologist at Pandora. The two men created a database that outlines a vast amount of musical components, the components that, as said before, Pandora uses to find for the listener new music that is similar in nature to the listener’s favorite song.
When a song is entered into Pandora, the system employs an algorithm that looks at these components for that song in question and then tries to find songs similar in terms of those details. For example, it examines the voice of the artist, especially the pitch, and the beat and syncopation of the music.
The workings of Pandora rely on graph theory, which is a branch of mathematics that looks at graphs, meaning lines and points. Musicologists at Pandora examine each song and rate each component. Then, every song in the Music Genome Project has its components’ ratings compared with every other song. Songs become connected, then; each song is on a node of a network. Edges forming between songs relate to how similar the two songs are, so then, songs that have stronger edges will show up more often in a listener’s radio station than songs with weaker edges. Pandora also employs the Strong Triadic Closure Property, which involves three nodes: if Node 1 is very similar to Node 2 and Node 3, then Node 2 and Node 3 must be slightly similar. Relating back to the music, then, Pandora will select songs similar and somewhat different to the listener’s preference so as to introduce the listener to new music.
All in all, Pandora and the Music Genome Project are incredible accomplishments in the fields of science, technology, and music.
Written by Aishwarya Vishwanath