Pandora Radio Technology

Music Genome Project

Will Glaser and Tim Westergren created the Music Genome Project to find a better way to analyze and categorize music. Previous “recommendation” services based their choices largely on genre similarity or popularity among other users, but the choices these types of services made were scattershot at best. The Music Genome Project developed a way of breaking each song down into its component parts, allowing a much more thorough analysis of musical content.

The heart of the Project is the concept of musical “genes”. Each gene is a descriptor of some element of the song, such as “electric guitar accompaniment,” “heavy syncopation” or “bluegrass instrumentation.” Rock songs tend to have between 150 and 200 individual genes, while more complex styles like jazz or classical can have up to 500 genes. Identifying as many genes as possible in a song allows for the most thorough analysis of its structure and content.

To analyze a song, one of the Project’s musicians listens to the song multiple times, identifying genes present and rating them between zero and five points. A five-point gene represents a theme or sound heavily represented in the piece, while zero represents a gene that is not present. The highest-scoring genes become “focus traits,” and these genes become the most important ones used in the matching process. Artist profiles contain genes that appear frequently in their music.

When you enter a song or artist into Pandora to create a station, the system first searches its database to see if a record exists. Once the system has a profile to work with, it scans through its database for songs with similar profiles, and queues them in the resulting musical channel. As you indicate your preferences while listening to the selected songs, Pandora uses that information to identify which focus traits are most important to you. When you approve a song, the traits included gain more weight in the system’s calculations, while rejecting a song reduces them accordingly. In this way, the system slowly alters the content of the channel to better suit your tastes.