U.S. Congress Pioneers Audio Search Technology
The U.S. House of Representatives is one of the earliest adopters of emerging technology aimed at searching through reams of audio and video with instantaneous results.
“The House wants to be able to digitally archive hours of audio and video testimony, and then quickly pick off content by keyword,” says Mike Daniels, sector vice president of Science Applications International Corp. (SAIC), San Diego.
Other end-users of this technology might range from military analysts on surveillance missions to law firms, political campaign organizations, TV broadcasters, and corporate marketers.
Multimedia search engines are nothing new, in and of themselves. Until recently, though, audio and video content needed to be laboriously viewed and “tagged” by humans before becoming searchable.
Now, however, companies like Atlanta-based Fast-Talk Communications Inc., a research start-up spawned at the Georgia Institute of Technology, are making it possible to perform automatic phonetic searches of digitized voice.
When you’re typing in a keyword for Fast-Talk to search, you can even misspell it, as long as the keyword is spelled like it sounds, Fast-Talk officials maintain. (For example, you might misspell Moammar Kaddafi’s last name as “Quadafy,” and still come up with the same results.) So far, Fast-Talk’s audio search engine is available in English and Spanish.
With Fast-Talk, accuracy supposedly improves as the search string gets longer. You’d get better results from typing in the phrase, “The bomb is about to explode” than from typing in either “bomb” or “explode.” With many text-based search engines, the opposite tends to be true.
SAIC, the systems integrator that’s implementing the solution for the House, recently made a $2 million investment in Fast-Talk to help “move the technology into the government and commercial worlds,” according to Daniels.
SAIC has also been integrating Fast-Talk with Vienna, Va.-based Convera’s Screening Room software, to combine audio search capabilities with closed captioning and/or speech-to-text conversion, natural language search, image-based queries, and high- and low-resolution clipping and republishing options.
According to Fast-Talk officials, the Fast-Talk search engine can find specific terms in recorded speech at a rate 36,000 times faster than real time. Specifically, 20 hours of content can be searched in one second, with 99 percent accuracy.
One factor behind Fast-Talk’s rapid results is indexing, a process that can be performed either during the recording period, or afterward. Alongside the discrete index file and digital sound track, the engine also builds a high-speed phonetic search track.
“The House will look next at using Fast-Talk for telephony,” Daniels says. In April, Fast-Talk unveiled a new product geared to searching voicemail and recorded phone calls.
Utopy, San Francisco, another player in the emerging audio search market, sells a solution called SpeechMiner, which is targeted at telephony applications within contact centers and enterprises.
