Search Inside The Music is a project of Sun Labs directed by Paul Lamere. The purpose is to skilfully link songs, by methods much more sophisticated than the usual tags genre or artist.

I.e. in a collection of music, it's possible to cross the genres to find 2 titles which have similarities of timbre or “musical energy”.

Concretely, the project explores new methods of music search thanks to the musical contents and the context. It's then a question of helping us to find and organize our music based on properties such as the words, the musical topic, the melody, the tempo, the rate/rhythm and the instrumentation. It is also possible to integrate the musical practices of the user into it.

The first thing to be made, it is to generate metadata directly from audio contents: pitch, harmony, timbre, instrumentation, tempo, rythm patterns, energy level.
That leads to a problem of resources. Sun Labs analyzes the musical properties frame by frame. Each frame is 40 milleseconds of music, which means 5000 frames on average per song, which means plenty of data to be treated.

Multiply that by the number of songs in an even modest collection, and one speaks then about million frame, a colossal calculation. This is a case where being part of Sun really helps. They have a “grid computing”, a wall to which it is enough to be connected to have the resources to thousands of network computers, as well as their storage capacities. According to Lamere, a simple computer would take 6 years to analyze a collection of 2 million songs, while same work on a grid of 1000 computers would take one weekend.

One of the most impressive functionalities of the system is the visualization of the acoustic distances between songs of all genre. The researchers of Sun Labs analyzed nearly 5000 songs, categorized by the genre specified by the artist, and encrusted in a 3D array according to their acoustic similarities.