Voice-recognition technology for the cockpit maturing fast

 - September 19, 2006, 10:36 AM

After being dismissed as impractical for cockpit use, voice-recognition technology appears to be getting a closer look from business jet makers thanks to recent advances.

Major business aircraft OEMs and avionics makers are keeping close tabs on voice-recognition research and development. The field has exploded in the last few years, with technology advancements finally catching up to the lofty expectations of developers and users. The latest generation fighters, including the Eurofighter and the Joint Strike Fighter, make heavy use of voice recognition for command and control functions, including weapons selection. The pilot still must push the fire button, but he can use voice commands to choose the type of missile that will do the job.

Experts agree that fighters and attack helicopters of the future will incorporate some form of voice-recognition technology, and some say it is only a matter of time before similar concepts end up in civil airplanes.

Cessna is exploring uses for voice recognition but mainly as a way to ease the paperwork burden of its quality-assurance crews who do the final check flights of airplanes before delivery. Instead of carrying clipboards and paper evaluation forms, check pilots could simply say aloud what discrepancies they have found and the voice-recognition software would transcribe those words into text and even assign them to the proper category.

When voice recognition will show up in the cockpits of business jets is hard to say. Honeywell and Rockwell Collins explored speech recognition during development of the Primus Epic and Pro Line 21 integrated avionics systems more than a decade ago, but at that time the technology wasn’t ready for prime time. Early voice-recognition software required a certain amount of training on the part of the computer software, which had to learn the nuances of the speaker’s speech pattern. The cockpit of a business jet in flight is also a relatively noisy environment, and background noise was the mortal enemy of early voice-recognition software.

Today’s software, however, is far more sophisticated and adaptable. Not only can modern voice-recognition systems instantly understand almost any speaker, they are even designed to decipher words if the speaker’s first language is German, Italian, French or perhaps something else. In fact, today’s software considers American English and British English two entirely different languages, and it is adept at understanding each.

Conversay, a developer of voice-recognition software for the mobile phone and computing industries, has turned its attention to aerospace in recent years. The company has competed on a variety of military projects, including the Joint Strike Fighter, and it is currently holding talks with players in civil aviation about bringing its concepts to the cockpit. Realizing that there are some potentially big hurdles related to the certification of voice-recognition technology in aircraft, the company has opted to start small by proposing simple voice-controlled menu software for electronic flight bag (EFB) tablet computers.

An EFB loaded with the Conversay software would interact with other applications, such as Jeppesen approach charts, in a way that would let a pilot choose charts using voice commands rather than touching the EFB screen. For example, a pilot could simply speak the words “Seattle-Tacoma, ILS, Runway 16 Left” and the appropriate approach plate would appear on the unit’s display. The entire time, the pilot could be scanning for traffic without having to look down until the chart was loaded and ready for use.

If Conversay and its partners are successful in bringing speech-recognition technology to EFBs, the company has a plan to expand into other flight-related areas, such as FMS programming and radio tuning. Conversay engineers say the software has the advantages of operating in extremely noisy environments. It also incorporates “out-of-vocabulary rejection” technology, meaning that the system’s microphone stays on all the time, yet it ignores spurious noises or conversations that might be going on around it. Once it hears speech that it decides is a command from the user, the software comes to life and begins performing selected operations.

“More and more of the instrumentation in military and commercial airplanes is computer-interface based,” said Peyvand Khademi, vice president of engineering for Conversay. “Having a full keyboard and mouse in the cockpit is cumbersome, and if, for example, you have a screen that has various menus and you try to use a joystick or cursor device, that’s hard to do, especially if you’ve got a multitude of other things you need to be doing. But if you can select a menu just by saying the label, that becomes a much more efficient way of completing the task.”  

Technology Breakthrough

Conversay engineers have worked alongside counterparts at Boeing and General Dynamics UK, and are about to start a project with Sikorsky, targeted at recent Air Force flight-test programs that are exploring speech-recognition concepts for pilot training. The company was also a finalist in the JSF contract bid and is currently involved in another bid for a contract related to a European fighter program.

On the civil side, voice recognition holds the promise of reducing workload and keeping the pilots’ heads up and forward rather than looking down at an FMS keyboard or EFB screen as they program waypoints and access menu pages.

“One of the areas where we see a natural application is when you have a single-pilot operation, for example in a small business jet, as a way to keep the workload manageable,” said Phil Moylan, a marketing and engineering consultant for Conversay. For a pilot flying IFR without the assistance of a crewmate, managing an EFB increases overall workload. Voice recognition presents an opportunity to free that pilot to concentrate on other tasks, the most important of those being the tasks related to flying the airplane. “If you could just talk to the EFB instead of looking down and selecting charts from a menu, the workload would be greatly decreased,” Moylan said.

An EFB loaded with voice-recognition software, in turn, could talk back to the pilot to verify that the software heard what he really said. The speaker’s commands would be picked up by the headset microphone, or possibly a microphone installed elsewhere in the cockpit. The software could decipher a Texan’s deep Southern drawl just as reliably as it could process any other regional dialect.

While early software was “speaker dependent,” meaning the user had to sit down and train the system by talking to it for about 30 minutes before it would pick up on all of the nuances of a particular voice, today’s software is “speaker independent” and “speaker adaptive,” meaning it can understand a broad cross section of people speaking English. It wasn’t until these breakthroughs that the promise of the technology could be fully realized, Moylan said.

“These were two features of voice recognition that had to happen before it was a practical solution for the aerospace industry,” he said. Speaker independent, he explained, means that the system needs no training for it to function properly. One crew could fly a particular airplane in the morning and another in the afternoon, and the speech software wouldn’t miss a beat.

A speaker with a voice anomaly, say someone who says “Seattle” as “Sheattle,” might cause a small amount of confusion for the system, but not for long. The software would analyze the word, and then ask, “Did you mean ‘Seattle’?” If the speaker replies in the affirmative, the software would recognize that word from then on. This is what is meant by the term speaker adaptive, Moylan explained.

He went on to point out that makers of modern fighter aircraft have requirements for voice-recognition technology that can be used in a noisy, supersonic cockpit, by a pilot wearing a mask and being subject to high g loads, and by aviators who will often be speaking English as a second language. Conversay’s current English-language software can recognize accents by speakers of nine foreign languages. Considering these demanding requirements, applying the concepts to business aviation should be a relatively straightforward process, at least in terms of the technology. Convincing the FAA is another matter, Moylan said.  

In the last three years, Conversay has worked to hone its software to satisfy aerospace requirements. Meeting the FAA’s standards for integrity in the avionics world is no small challenge, but the company has invested significant resources to make its speech engine an “FAA certifiable” product, Moylan said.

And while a few business jet OEMs have expressed interest in applying voice recognition to the FMS, he said that nearly everyone involved recognizes that the certification path will be a slow and difficult one. “The FMS would be a really great target,” he said, “but you don’t want to start with that one.”