Military seeks up-armored speech translator

DARPA project aims to pick up voices in noisy conditions

It's interesting how the military often creates complex names for products or programs it develops whose acronym turns out to be funny. I think everyone's familiar with the origin of the word "snafu."

So it is with the Defense Advance Research Projects Agency (DARPA), which this month publicized an invitation for someone, anyone, to develop technology that does Robust Automatic Transcription of Speech, or RATS. I can just imagine one soldier in Afghanistan interrogating a foreign-speaking suspect telling his buddy, "Let's turn the RATS on him."

But that just might happen. Under DARPA's Broad Agency Announcement process -- or BAA (not to be confused with the sound a sheep makes) -- the agency is looking to develop a device that does three things, all well: accurately recognize speech; identify the language and the speaker; and work in noisy environments.

Although speech translation devices have already been developed, and some of them used by coalition forces in Iraq and Afghanistan, the DARPA request is for something hardened for the battlefield. DoD wants the Humvee of speech translators.

"Existing transcription and translation and speech signal processing technologies are insufficient for working with noisy or degraded speech signals that are of importance to current and future Department of Defense (DoD) operations," DARPA explains in a 36-page document describing the project.

Languages the system needs to understand are Arabic, Farsi, Pashto, Dari and Urdu.

From the description in the document, it looks like the DoD is looking for something more than the handheld translators sold by companies like Ectaco or the software programs coalition forces ran on laptop computers. Ectacto, by the way, also sells a translation program that runs on Microsoft Pocket PC devices. And a group inside Microsoft Research is dedicated to improving speech recognition software.

All that aside, it seems like the DoD wants RATS to recognize speech picked up through eavesdropping equipment or through monitoring of radio transmissions. Capabilities the system needs to have, the DoD says, include "Speech Activity Detection: The ability to determine whether a signal is actual speech or background noise, or music."

The document also emphasizes the importance of hearing accurately what people are saying: "The United States has a compelling need for reliable, relevant information to directly support intelligence gathering in the field, to inform military decision makers and respond to national security requirements."

In other words, they don't want any snafus.

Michael Cooney delves into the RATS project further on his Layer 8 blog.

• Microsoft