Search Inside the Music from Sun

Sun's Search Inside the Music

For those of us who love music, wishful thinking always includes the creation of a playlist that's specifically tailored to our interests and mood. Current recommender systems utilize one of two ways to sort music: collaborative filtering or content matching. Amazon uses collaborative filtering and offers suggestions based on what other's have found interesting. Pandora, as it turns out, uses content matching by hiring musicians to analyze songs and group them by hand into micro-genres. The Search Inside the Music poject from Sun aims to bring together the best of both worlds!!

Being the greedy individuals that we are, the best approach is one that offers the best of both worlds. Enter, supposedly, Sun's Search Inside the Music project! This project, led by principal investigator Paul Lamere, is is believed by team members to be the best music similarity algorithm because it's based on the actual sound. Leading recommenders categorize music based on artist, album, song and genre. Search Inside the Music analyzes features such as frequency, beats per minute, pitch, harmony, key, timbre, instrumentaion, tempo, intensity and energy level to map out the rhtyhm structure and determine the genre and which instruments are playing. In essence, it uses acoustic similarity to help people find music that "sounds similar" to music that they already like. According to Lamere, it takes a Pandora staffer about 20 minutes to categorize each song. The Sun system, on the other hand, is fully automated -- it's a computer-driven algorithm capable of whipping along at a clip of 3 seconds per CPU per song.

One of the more impressive capabilities of the Search Inside the Music system is its ability to generate a visualization of the acoustic distance among songs of different genres. This display shows which specific songs are similar to other songs of the same genre, but it also illustrates the degree of acoustic similarity between songs of different genres. This type of visualization provides a way to quickly create customized playlists based on acoustic similarity.

"For examples, let's say you've had a rough day at work; you're leaving the office and heading into heavy rush-hour traffic, and you want to hear myusic that will help you reduce your stress level as you drive home. The Search Inside the Music system can quickly generate a playlist that serves as a 'musical journey,' starting with higher-intensity songs that match your current stress level, and gradually diminshing the intesity of the songs as you make your way home. So during your 50-minute commute, you make the transition from Rage Against the Machine and led Zeppelin to Schumann's piano music--smoothly and seamlessly. And all the songs that are played are songs that you like."

Sun's other innovation is a tagging system that categorizes music based not on who's purchased it but on its attributes, described with tags like "quirky", "indie", "rock", "fast", "frenzied", "90's", or "cute" and "fun". For example, querying Sun's prototype search engine for Led Zeppelin brings up "tagomendations" such as the Rolling Stones and Jimi Hendrix. The user can then click a "why" button to find out why a particular song was recommended. Hendrix is recommended for Led Zeppelin based on tags like 'guitar gods', 'classic rock', 'guitar virtuoso' and 'psychedelic'. Sun is compiling these tags by searching reviews, lyrics, music blogs, social tagging sites and artist biographies, and incorporating the information into a prototype search engine. Compiling the tags based on a comprehensive search of the Web prevents people from gaming the system by generating their own tags to enhance the popularity of certain tracks.

In addition to recommendations for other music, the search engine provides links to videos, pictures and upcoming concerts, if the artist search for is alive and touring.

While there are certainly aspects/feelings that you can't get from analyzing the audio, it can help artists who are so new that they fly under the radar.

The more prevalent concerns about such a system stem from scalability. Where would analytical data be stored, how can the system be scaled to handle billions of songs and how can we find the computer power necessary to support such a system. Sun's computing model called grid computing, which lets you plug into a huge network of extremely power computers and draw on their combined CPU power on demand, provides a promising solution. Future advancements to the computerized analysis will enable recognition of major and minor chords, bridges and choruses, and the rhythm patterns of reggae, pop and ska according to Lamere. Though the Search Inside the Music project is not likely to be released as a commercial product, Sun officials say they plan to make the software available as open source, perhaps within six months.

A new music recommendation system from Sun (November 5, 2007)
Machine learning fuels Sun music recommendation technology (October 31, 2007)
Sun Micro spins its music software (April 9, 2007)
Search Inside the Music (September 26, 2006)

Emerging Recommendation Technology Helps Pick Shows

"Imagine you're listening to the radio in your bedroom. A song comes on that catches your ear - let's say "Ayo Technology," a hit single from the latest 50 Cent album. It's just the kind of jam that puts you in a good mood, but when it's over, a tinge of disappointment sets in. Surely the next song won't measure up. But what if it did?".........

Possessing an uncanny ability to relate to the sentiment above, I'm left to ponder what if the radio could predict what song I would enjoy hearing next....what a joy it would be to listen to all my favorite songs without ever having to skip a single one!!! And what if this same concept applied to movies, online games, and television shows alike! Apparently that is the direction recommendation technology is heading.

Recommendation Technology attempts to make educated guesses as to what else may interest you and is an extremely useful mechanism for any company selling items or any company that has a platform to distribute items. Companies have started realizing that perhaps the most effective marketing tool lies in their ability to personalize "suggestions" and tailor these recommendations to the interests of each individual. Given the plethora of information available via the internet, it can be challenging to find exactly what you want to find and as studies suggest, sometimes you don't know exactly what you're looking for which can amplify the problem.

While recommendation technology is not new, it has not been very effective in the past. As inferred in the "My TiVo thinks I'm gay" plot, perhaps recommender systems attempted to be too intuitive and less personalized initially. After recognizing the benefits available by offering more accurate suggestions (increased sales and session length) and the consequences seen by offering inaccurate suggestions (less sales, lower likelihood of returning), systems are seeking to become more sophisticated. "When you get down into the deep mathematical analyses of this stuff, and you have good data on content and users, you can make good predictions of what users will like."

Even more interesting is the fact that personalization isn't limited to just the web. Researchers believe that more than 1/3 of daily television viewing will be on-demand in the next 5 years. As an advocate of personalization, Im intrigued to see what other areas can be explored beyond the internet and television...

Emerging Recommendation Technology Helps Pick Shows