nfotainment. The very term conjures up images of buttons, knobs and display panels. But experts say that if automotive infotainment systems are to reach their full billion-dollar market potential during the next few years, the key to success will not be a glitzy instrument panel, but rather voice-recognition software with the ability to filter out the hum of a tire, the whistle of the wind and the din of a radio while deftly picking out the subtleties of human speech.

According to experts, voice recognition is a non-negotiable component of automotive infotainment systems for a simple reason: It will help drivers avoid accidents. Automotive experts and legislators fear that a profusion of electronic gadgets in automobiles will distract drivers. "People can cope with a lot of distractions," said Phil Spelt, a senior research scientist at Oak Ridge National Laboratory, which is conducting a soon-to-be-released study on driver distraction for the U.S. Department of Transportation. "But all it takes is one split second when they can't cope, and you have a major accident."

Given such concerns, it's no wonder that industry analysts are predicting a sharp upswing in the use of voice-recognition software in automobiles. They expect voice recognition to serve as an interface for car radios, CD players, navigation systems and cell phones. Used in conjunction with speech-to-text and text-to-speech programs, experts also see it as a means to send and retrieve e-mail, as well as connect to the Internet.

A recent study by Cahners In-Stat Group (Scottsdale, Ariz.) estimated that the sale of speech-recognition systems for telematics, handheld devices, appliances and other applications is expected to reach $2.7 billion by 2005. Researchers who conducted the study said that automotive applications for voice recognition could explode, especially in one key area. "You can only make so much money with a speech product that adjusts your radio or your air conditioner," said Brian Strachman, a senior analyst for Cahners In-Stat. "But if you're talking about connecting your car to the Internet and to the outside world, there's a huge voice-recognition market for that."

Work to be done
The drive to use voice recognition in automobiles is hardly a new one. General Motor's OnStar division already employs it in its Virtual Advisor, which is expected to reach more than a million vehicles. Clarion's AutoPC also uses it, and a speech-recognition system from Visteon Corp. (Dearborn, Mich.) has made a media splash in Jaguar's S-Type vehicle.

Under ideal conditions, makers of such systems say they can be extremely accurate. "We can consistently deliver 97 percent accuracy under certain conditions," said Scott Pyles, director of product management for automotive products at Lernout & Hauspie Speech Products NV (Ieper, Belgium). "But in a dead quiet environment, you'll get a different accuracy level than you will at 30 miles per hour with the windows open."

To achieve speech recognition's full potential, next-generation software programs will need to deal with a wide variety of noise sources, including wind, road, defroster, fan, windshield wipers and back-seat occupants. They'll also need to tangle with less obvious problems, such as line echo, electrical interference and signal strength.

Such problems reportedly are far more manageable at 30 mph than they are at 70 mph. In particular, engineers say that road noise, wind noise and even back-seat occupant noises become magnified at higher speeds. Some engineers believe that accuracies at high speeds drop as low as 70 percent. "In terms of vehicle noise, the real killer seems to be the defroster in winter and the air conditioner in summer," said one automotive engineer. "The fan is a problem all year round."

Lernout & Hauspie engineers say they are dealing with noise problems by developing specialized routines for noise cancellation and acoustic echo cancellation. "We design the speech engines by collecting a great deal of information for noise models, language models and word models," said Pyles of Lernout & Hauspie. "We collect data in vehicles with the windows up, windows down, on cobblestone roads, on dry roads, on muddy roads, with the fan on, with the fan off, with the radio on and with the radio off." The result, Pyles said, is that the company's software improves each year.

Some software makers, such as Clarity LLC (Troy, Mich.) and Conversational Computing Co. (Redmond, Wash.) say that the solution to the noise dilemma lies in the use of specialized software. Clarity, for example, offers a technology known as Clear Voice Capture, which extracts the voice signal of interest. The company says that the technology provides an improvement over noise suppression systems, which have difficulty with signals that have components overlapping voice signals. Clarity's product uses time-domain signal-processing techniques to extract the audio signal of interest from a mixture of sounds, including background noise and music.

Some companies have also tried dual microphones and so-called "array" microphones. Most say, however, that array microphones, which cost between $100 and $180, are currently outside the acceptable cost limit for such automotive applications.

Divergent paths
Automakers currently are split on whether speech systems are ready for prime time right now. General Motors and DaimlerChrysler both use speech technology today. Vendors, too, have rolled it out in new products: Delphi Automotive incorporated it in its new Communiport system and Visteon put it in ICES (information, communication, entertainment system).

BMW and Volvo, however, are not as committed to it in future plans. BMW recently rolled out iDrive, an infotainment system that allows users to change functions by pushing, pulling or twisting a console-mounted knob. Similarly, Volvo is said to be planning an infotainment system that employs steering-wheel-mounted buttons. And Johnson Controls, a major automotive supplier, has said that it plans to use "voice recognition only as an overlay."

Many developers of infotainment systems, however, can't imagine selling systems that don't rely on speech technology. Any interface that doesn't abide by the philosophy of "eyes on the road, hands on the wheel," is an invitation to distraction, they claim. "For complicated functions, speech technology is superior to buttons and knobs," said John Slosar, director of telematics multimedia products for Visteon. "We don't expect people to drive cars and read screens at the same time. We think text-to-speech and speech-to-text is the right way to minimize distractions."

Ultimately, experts believe that voice recognition's automotive success is not so much a matter of if, but when. As speech engines improve, most believe that voice recognition will take on a bigger role in automobiles. "The consumer is becoming more aware that he doesn't always need to push a button, twist a knob or flip a switch," Pyles said. "As speech technology becomes more natural-language-oriented, consumers will grow more comfortable with it, and then you'll see it really take off."