Clio: the algorithmic Music Genome Project
May 18, 2011

Orpheus Media Research working on a music discovery platform based in machine learning

http://vator.tv/news/2011-05-18-clio-the-algorithmic-music-genome-project

Technology trends and news by Ronny Kerr

If you’re even a little tech-savvy, you’ve probably heard of the Music Genome Project, the technology that powers all of Pandora’s radio stations. The Project aims to break music down to its basic “genes,” an effort intended to find those fundamental elements that make a pop song sound like pop and a jazz tune sound like jazz.

What you probably didn’t know, however, is that the Music Genome Project requires literal manpower to classify all those tracks in Pandora’s library. That’s right: “seasoned, highly-trained analysts” sit down with every single song, one by one, categorizing them by way of hundreds of different musical attributes, from structure and composition to lyrical content.

As you can imagine, having real, live human musicologists analyzing music might be a great way of building an intelligent music recommendation system, but it’s not really scalable. Consider that Pandora currently has 850,000 songs in its system, versus the 10 million songs on Rhapsody, a music subscription service.

In direct contrast to Pandora’s approach, let me introduce Orpheus Media Research (OMR), a Brooklyn-based music research and development company responsible for Clio Music, a music discovery platform that sounds a lot more like IBM’s Watson but for music.

Last week, I got a chance to talk on the phone with two people leading the charge at OMR: Greg Wilder, founder and Chief Science Officer (amazing title), and Alison Conard, COO.

“We’re talking about machine learning and algorithms,” says Wilder, who describes himself as a Pianist/Composer/Theorist-Turned-Entrepreneur/Technologist.

Naturally, I had to ask whether Pandora maybe had an advantage in using people to classify songs in its library. After all, wouldn’t you rather have a person, not a computer, recommend a song to you?

Wilder, however, insists that the brilliance of Clio is that it actually “listens to elements in a song that a person would listen to. There are limitations in our own brain that help us listen to music and distinguish different artists,” and Clio’s technology exploits those limitations.

What is the melody doing? What are the melodic hooks that make the music recognizable? What is the bass player doing? How is the drum machine or drummer affecting the mood? What specific grooves does the rhythm section use?

And on and on.

“We haven’t been able to trust computers yet because it hasn’t worked yet,” said Conard. “We have a system we believe people will be able to trust.”

Clio just launched, but it builds on Myna Music, a platform that beta launched in the fall. In terms of application, OMR is first looking to the production music industry, a “shadowy world” that Wilder tells me accounts for approximately a third of the entire music industry. Think music in TV commercials, sitcoms, movie trailers and advertising, and you instantly know what he’s talking about.

Right now, finding the right song involves matching a search with manually-entered keywords, like “romance” or “drama.” Clio is much more intelligent than that.

The potential use for Clio, though, is unlimited. Though it isn’t a consumer-facing service itself, tons of emerging Web music services–MOG, Rdio, Slacker, Google Music, even Pandora–might all be using Clio’s music discovery technology some day.