Tools help pinpoint the right words

In advance of speech transcription research breakthroughs, intelligence organizations could boost their analysis capabilities by using audio-mining technologies already on the market to narrow down the amount of material they need to review, experts say.

Those technologies create indexes of audio files using words or speech sounds and enable users to search those files in ways similar to how search engines probe the Internet.

"We don't see ourselves replacing the existing world of speech-to-text in agencies, [but] we think we make the system more efficient and effective," said Patrick Taylor, vice president of sales and marketing at Fast-Talk Communications Inc., based in Atlanta.

"Our intelligence agencies...have a triage problem. They've got more coming in than they can put through their [speech-to-text] systems," Taylor said.

Fast Talk's technology could help sift through that data, he said. One example would searching for 1,000 key concepts, such as Osama bin Laden or al Qaeda, and then drawing the results to analysts' attention.

Fast Talk builds an index of real-time, prerecorded speech from phonemes, or fundamental speech sounds, Taylor said. Users then search the index using a keyword, which is converted into a set of phonemes. Search results are ranked according to how well they match the search term.

Twenty hours of recorded speech can be indexed in two hours and searched in a second, Taylor said.

Since it uses phonemes rather than words, the system can be searched "using a natural language approach that doesn't care how you spell the stuff," said Greg Pepus, visionary solutions architect at In-Q-Tel, the CIA's venture capital firm, which is evaluating the technology.

Fast Talk is lining up federal customers for the technology, which is available in English, Spanish and Arabic versions, Taylor said. A basic system with 25 server seats costs $60,000.

Another possible solution is ScanSoft Inc.'s audio-mining tool, which creates searchable audio file using keywords or concepts.

"Based on a recorded set of phone conversations, [for example], you could search for specific keywords that might flag a security risk," said Robert Weideman, vice president of worldwide marketing for ScanSoft.

"Instead of listening to every conversation in sequence, [you] would only have to home in on a specific conversation that matched" your search term, he said. "It would then take you within the conversation to the precise location where that match was found."

The technology is used for screening telephone calls at some prisons, and the company is eyeing potential federal intelligence customers. The tool is only available in English but could be adapted for other languages, Weideman said.


  • Defense
    Ryan D. McCarthy being sworn in as Army Secretary Oct. 10, 2019. (Photo credit: Sgt. Dana Clarke/U.S. Army)

    Army wants to spend nearly $1B on cloud, data by 2025

    Army Secretary Ryan McCarthy said lack of funding or a potential delay in the JEDI cloud bid "strikes to the heart of our concern."

  • Congress
    Rep. Jim Langevin (D-R.I.) at the Hack the Capitol conference Sept. 20, 2018

    Jim Langevin's view from the Hill

    As chairman of of the Intelligence and Emerging Threats and Capabilities subcommittee of the House Armed Services Committe and a member of the House Homeland Security Committee, Rhode Island Democrat Jim Langevin is one of the most influential voices on cybersecurity in Congress.

Stay Connected


Sign up for our newsletter.

I agree to this site's Privacy Policy.