Air traffic trainees hear new voices

Simulations put students of air traffic control into highly realistic situations, complete with the voice of a human pilot.

Courtesy of Adacel

Air Force trainees are learning air traffic control maneuvers with the help of software that listens to their instructions and responds like a human pilot.

In a $72.5 million deal, the Air Force has deployed two Tower Simulation Systems out of the 94 that it plans to install by 2007 at all bases, including Air Force Reserve and Air National Guard sites. Last month, the Air Force exercised its option to purchase 46 of those 94 tower simulators, valued at $23 million, or almost a third of the total deal.

The contractor, Adacel Systems Inc., is an arm of Adacel Technologies Ltd. of Australia.

Eglin Air Force Base, Fla., and Barksdale Air Force Base, La., are testing the simulators now. Next to receive them are air traffic control training centers in Little Rock, Ark., and Grand Forks, N.D.

Adacel's MaxSim software offers third-party voice recognition and text-to-speech technologies. 'You can walk in and begin speaking to the system, and it will respond,' said Tom Harris, tower simulator program manager at Wright-Patterson Air Force Base, Ohio. 'There's no voice training. There's no building voice files.'

Trainees plug their headsets into a console with a 270-degree simulation of a runway with departing and incoming planes. Software algorithms analyze the trainees' acoustic signals and turns them into words.

A separate module for pilot behavior then applies more algorithms and probabilities to select the best meaning from a precoded list'for example, 'Sierra two-six, you are No. 3 to land. Follow three-eight-five on downwind, over.'

The system 'doesn't have to just recognize what the student said,' Adacel product manager Gary Pearson said. 'It has to recognize what he meant.'

It then modifies the visual display and sends an appropriate synthesized answer to the student's headset.

No more tabletop landings

The simulator uses application programming interfaces to separate modules running under Linux and Microsoft Windows XP and NT operating systems across several servers. It will replace the Air Force's previous training scheme of dragging model airplanes'known as 'airplanes on a stick''down a tabletop under the eyes of an instructor.

With the simulator system, 'You can put the student in a real-life environment,' Harris said. It removes the need for so-called pseudopilots, or instructors, to call out pilot responses to the students over an intercom.

Adacel's visual graphics and image generators come from Quantum3D Inc. of Orlando, Fla. Its voice capability stems from a combination of a Voice Extensible Markup Language grammar and vocabulary builder from Newfound Communications Inc. of Lawrence, Mass., and Nuance 8.0 speech-recognition software from Nuance Communications Inc. of Menlo Park, Calif.

The text-to-speech technology, which assigns a human voice to words, comes from Rhetorical Inc., a U.S. subsidiary of Scotland's Rhetorical Systems Ltd.

The text-to-speech software breaks down hours of speech recorded by hired actors into individual sounds and intonations stockpiled in a database.

It reconstructs the sounds to form words, also stored in an electronic lexicon and without the pauses or monotony of most synthetic voices. The voices range from those of a Scottish male to a California valley girl.

The Air Force will use American male and female accents rather than British or Australian ones. 'The computerized voice can sound indistinguishable from the original voice,' said Marc Moens, Rhetorical's chief executive officer. 'It actually sounds like real people.'

Stay Connected

Sign up for our newsletter.

I agree to this site's Privacy Policy.