Speech Recognition Makes It into the Mainstream


    The following passages are excerpted from "Electronic Engineering Times", and suggest a major push into automotive speech recognition systems next year (2001). It will probably be years before automotive speech recognition trickles down to everyone, but there will be a developmental commitment underway that can only force-feed the technology.
 

"Ford, GM go for car talk technology

                     By Charles J. Murray
                     EE Times
                     (04/21/00, 11:47 a.m. EST)

"PARK RIDGE, Ill. Stepping up the race to bring e-mail to vehicles, two of the world's biggest carmakers have formed partnerships to develop better voice-recognition systems.

"Ford Motor Co. will invest $20 million in Lernout & Hauspie to develop speech technologies while General Motors is teaming with General Magic Inc. and Nuance Communications to continue work on voice-recognition systems. In a separate move, Delphi Automotive Systems has allied with Lernout & Hauspie to put voice recognition in a Palm V handheld computer and docking station that links cars to the Net.

"The race to create more effective speech systems is now seen as critical for automakers. Several of them, most notably Ford and GM,  have said they are devoted to an "eyes-on-the-road, hands-on-the-wheel" philosophy as they work to incorporate e-mail capabilities into automobiles.

"That's why OnStar announced two weeks ago that it will offer its so-called Virtual Advisor in 2001 General Motors automobiles. Virtual Advisor will enable drivers to access stock quotes, news headlines, sports scores, weather and e-mail through their cellular phones and voice-recognition systems. The system uses a virtual voice with a vocabulary of several thousand words located at a remote Java-based server. It also employs an on-board voice-recognition system with a very small vocabulary just big enough to access the phone and connect the user with the server. Once connected, subscribers will go to the OnStar Web site, where they can link up with services.

"It's almost like one huge radio station," Peterson said. "But instead of a smattering of information, users will get personalized services." OnStar expects the Virtual Advisor to be in a million vehicles by year's end, a spokesperson said. OnStar's design philosophy differs dramatically, however, from that of Visteon Corp., Ford's vehicle systems and component manufacturing operation, and Clarion Corp., which offers an aftermarket in-car PC. The key difference, say engineers, is that OnStar's system is "thin" meaning that it incorporates few on-board electronics. A basic processor board using Windows CE carries only 12 Mbytes of memory just enough to handle a small vocabulary. Though OnStar won't identify the processor it's using, it will say that the processor provides enough capability to run 'Veronica,' the virtual voice that OnStar has used since its inception.

"A Web server for OnStar's Virtual Advisor uses more than 700 Mbytes of memory and a 1-GHz processor. "Voice-recognition capabilities are a function of a server's processor speed and memory," said Karen-ann Terrell, director of e-Vehicle Product Management for e-GM, a new General Motors division that develops e-business. "The reason we do it that way is that we can still add applications and keep cost out of the vehicle."

"To augment the voice-recognition system developed in-house at General Motors, OnStar forged the partnerships with General Magic (Sunnyvale, Calif.) and Nuance (Menlo Park, Calif.). General Magic, which is partially owned by GM, develops the VUI used in the server-based voice-recognition system. It then licenses voice-recognition technology for use in the system from Nuance.

"In contrast, Visteon's system is considered a "thick" client technology: Its processing power is located in the vehicle, not a remote site. Using the company's dashboard computer ICES (for Information, Communication, Safety and Security), Visteon can link drivers to the same news, weather, stock quotes and e-mail as OnStar. It can also link up to Visteon's RESCU (Remote Emergency Satellite Cellular Unit) for turn-by-turn navigation.


Ford's System uses an onboard Pentium 166
to provide news, sotck quotes, and e-mail.

"The design difference, however, is that ICES takes significant on-board processing power. It uses a 166-MHz Intel Pentium processor with a Windows CE operating system. "We know that the applications we launch today aren't the end-alls," said Lori Markatos, product team leader for ICES. "And we don't want new applications to bring the system down. So we essentially have a desktop in the car."

"Something ventured

"Visteon plans to develop its system further through a 40 percent stake in a new partnership with Lernout & Hauspie (Burlington, Mass.). expected to employ 50 people developing speech technology.

"Ultimately, the goal at both OnStar and Visteon is to bring out speech-recognition systems that will let users speak at normal speed, without needing to memorize a list of commands. That ambitious job requires the expertise of outside developers."