Sun Microsystems is expanding its horizons to include some search and indexing technology to complement its server market, primarily in the realm of audio search.
As companies and laymen become increasingly overwhelmed with the sheer volume of digital media, Sun Microsystems, in its California labs, is searching for a way to help people organize it. As Kevin Spencer Brown reports, the server company has three projects running-two delving into the intricacies of audio search, and one focused on automatic contextual organization.
“Imagine an iPod with a million songs on it,” said Sun researcher Paul Lamere. “Trying to scroll through a million songs, you’ll never find you want. Hit shuffle play, and you’ll never hear what you want.”
This is exactly the type of problem Lamere and Sun are trying to address with some nifty next generation artificially intelligent audio search technology. The search technology Lamere is concentrating on analyzes the acoustics, tempo and melody of different types of music and matches those qualities with similar types of music in order to make recommendations to users. Or, if a music fan is in the mood for something else, the system can recommend music with very different qualities.
For example, if a music lover exhibits a love of “jam bands,” the system can recognize certain properties of Dave Matthews Band songs and match them up with, say, Jason Mraz or Phish.
Also on Sun’s audio search burner is software designed for searching recorded speech, which “takes note of all the things the speaker might be saying.” Coupled with a timestamp, the software allows users to quickly search recorded speech with quick accuracy.
Finally, the third project, led by Steve Green, aims to analyze large sets of documents and place them into appropriate folders without guidance from human indexers. The system, once finished, would be able to identify new documents on various subjects and file them with older documents.
For example, if the text of a document is about hamburgers, the software will recognize important keywords like “bun” and “beef,” and know that it belongs in a food folder.
“When it encounters a word that might fit into multiple folders-say Mustang’-it looks for other shared words for a clue. Ford’ is less likely to show up in a paper about horses,” writes Mr. Brown.
The three technologies are primarily designed as server applications for corporate use, but the development of such systems will surely set Sun servers apart in the industry while influencing search capabilities elsewhere.
