CHAMPAIGN – David Tcheng has played in local bands for 20 years. Now he's putting his love of music and computers together to help iPod and MP3 player users organize their music collections.
Tcheng, a research programmer at the National Center for Supercomputing Applications, worked to develop software that can analyze music and categorize it.
Using "artificial ear" technology, he and his collaborators came up with One Llama, a system that not only helps listeners sort their music collections, but also recommends new songs they're likely to enjoy.
"What makes us unique is, our technology listens to a song, creates a data set (from it) using data mining and determines which songs sound similar," Tcheng said.
A "beta" version of the system is expected to be available for download later this month. People who want to try it can enter their e-mail address at One Llama's Web site – www.onellama.com  – and they'll be notified by e-mail when the system can be downloaded.
One Llama Media, the company commercializing the technology, has already attracted a $250,000 investment from IllinoisVentures, the University of Illinois-affiliated office that helps start-up companies get off the ground.
Rob Schultz, an IllinoisVentures senior director who is providing interim leadership for One Llama, said the market is ripe for a technology like this one.
According to Schultz, 188 million MP3 players and 200 million enhanced cell phones were expected to be sold in 2006. Sales of digital music are projected to soar from $1.1 billion in 2005 to $4.6 billion in 2008, he added.
With so many music choices available, consumers need a truly good "recommendation engine," he said.
As it stands now, listeners are overwhelmed by the choices. According to iTunes Registry, the average iPod user has 3,542 songs in his collection and actively listens to only 23 percent of them. Sixty-four percent of the songs are never played.
Several music recommendation technologies are on the market today, with Pandora probably the most popular. It's the product of Pandora Media, which was created by the Music Genome Project.
Pandora evaluates songs for 400 attributes, with musician-analysts listening to the songs and manually scoring them for each attribute, Schultz said. But Pandora focuses only on popular songs, limiting the music that can be recommended.
Also, because musicians do the scoring, "human error" can be introduced by variability in interpretation, Schultz said.
One Llama's approach eliminates human error by using software to analyze musical attributes of a song, such as melody, harmony and rhythm.
It extracts 3,000 features from each song and uses "advanced machine learning techniques" to measure similarity between songs, according to a presentation prepared by the company.
The system eliminates popularity bias and is scalable to large collections of music, the company states.
Another popular music recommendation technology is Last.fm, Schultz said.
"They're a European company doing recommendations, but they're not doing feature-based recommendations," he said, noting that Last.fm makes use of collaborative filtering.
Pandora makes feature-based recommendations, but "Pandora does it manually, and we're using the computer to do the feature extraction and analysis," Schultz said.
Plus, One Llama builds "user-defined" channels by modeling the similarity between songs selected by the user. A One Llama user tells the system which songs he or she likes and doesn't like, and the feedback helps One Llama refine the model.
Schultz said in the coming year, One Llama plans to get feedback from users to help determine what features should be added to the system. The company would then like to roll out a mobile version, but Schultz doesn't expect significant revenues until 2008.
"If we can get customers using the system, then investors will come," he said.
Schultz said the business expects to derive revenue from three sources:
– Commissions on songs One Llama recommends and the user goes out to buy.
– Monthly subscription fees for One Llama service on mobile phones.
One of Tcheng's collaborators, Kris West, plans to join the company in May. Also working to develop One Llama are Amit Sudharshan and Nikhil Pandit, both of whom studied computer engineering, and Chad Olson, a consultant who helped develop One Llama's user interface.
Michael Welge, a senior researcher at NCSA and a developer of "music to knowledge" (M2K) technology, has been an adviser to the company.