U.S. researchers turn words to music

Electrical engineers and computer scientists at University of California, San Diego, have developed "Google for music," a search engine that takes words and finds tunes to match.

U.S. electrical engineers and computer scientists have developed "Google for music," a search engine that takes words and finds tunes to match.

Gert Lanckriet and colleagues have developed a computer system that allows users to find music using words, such as 'funky guitar solos.' ((University of California, San Diego))

Users can search "high energy instrumental with piano," "funky guitar solos" or "upbeat music with female vocals" and locate the songs they want, said the researchers, from the University of California in San Diego.

They have built a system that allows a computer to annotate a song using algorithms they created. Once annotated, a user can retrieve it with a text-based search engine, they said in a statement Tuesday.

But it's not quite that simple. Before the computer can categorize songs, it has to be trained. Gert Lanckriet, a UCSD electrical engineering professor, said that the scientists have developed what they call Listen Game, which can capture the word-song combinations necessary to train the system to label large numbers of songs automatically.

Part of the problem inassigningattributes to songs is human: "A pre-teen girl might consider a Backstreet Boys song to be 'touching and powerful' whereas aDJ at an indie radio station may consider it 'abrasive and pathetic,' " the authors wrote in a paper presented in Vienna this week.

Listen Game helps address that. Theonline gamegets players to tag songs with words, determining which words are most — or least — relevant to the song. Players earn points when they pick the same words as others listening to the same songs.

The researchers use these"semantic annotations of music" to train their computer to accurately label songs it has not previously encountered, they reported in the paper.

"We've shown — in academic terms — that our game works. We're close to the performance we get with comparable survey data from the music undergrads we paid to fill out music surveys," said Doug Turnbull, a computer science PhD student who wrote the paper withLanckriet.

The researchers presented three papers at the International Conferences on Music Information Retrieval. As well as the one on Listen Game, they reported on identifying words that are most likely to be meaningful to music search engines, and detecting boundaries within songs.

Boundary detection is important forgenerating music thumbnails, efficient browsing and music information retrieval.

"Maybe you want to listen to the Beatles, but mellow Beatles. You don’t want to listen to Back in the USSR," said Turnbull. "We are building a system that lets you use natural language to search for music with this level of detail."

Listen Gameis free and has no advertising. An updated version is planned.

"A bunch of engineers made Listen Game. The second generation will have a livelier look and feel," Lanckriet said.