Freeman, the company has developed different versions of its software tailored to a variety of existing in-car systems. To develop its integrated voice recognition capabilities, VoiceBox has been working with companies like I.B.M., Toyota and XM radio. And quickly changing topics from “What’s the traffic like on I-95?” to “Turn it down!” did not confuse the system.
King was the artist.Īfter replying to a series of questions about current basketball scores, in response to, “What about Boston?” the VoiceBox computer also correctly surmised that we wanted to know about the Celtics and not the rock band or traffic in the city. Asking the system “Who’s playing?” brought a response that B.B. Simply saying, “I’d like to listen to jazz and blues,” turned on the XM satellite radio and tuned in the correct station while we hunted for a parking spot.
In a recent demonstration - driving around Manhattan with a prototype portable VoiceBox system - the advantage of such sophisticated software was immediately apparent. These digital agents are designed specifically to interpret the driver’s intention and determine the correct context for a given request. So VoiceBox developed specialized software programs that overlay existing speech recognition engines like those from I.B.M. “Having consumers memorize 1,000 different commands for the right context can be distracting, even deadly” if there is confusion when you are driving 60 miles an hour, he said. VoiceBox has been working on what it calls a conversational speech engine for cars that should be ready in the next year or so, he said.
The main problem in seamlessly switching between systems - entertainment and navigation, for instance - using conversational speech has been getting the car to understand the context of what the driver is saying, according to Tom Freeman, co-founder of VoiceBox Technologies of Bellevue, Wash. So why is a car not able to tell the difference between a request for a tune by Steve Winwood’s former band and the need to avoid congestion on Interstate 95? has developed for cars, for example, “is now as powerful or more powerful than ViaVoice was on a desktop computer” just six years ago, he said. Research in Yorktown Heights, N.Y., said. Did you want music from the ’70s supergroup or information about gridlock on your route?Īs costs have come down and the complexity of onboard computers has ramped up, the computing limitations are largely fading away, Roberto Sicconi, director for speech technology integration at I.B.M. Asking for “Traffic, please” at the wrong time can flummox most systems.
While such systems do not need to be trained like the dictation programs used on desktop PCs, drivers often find they have to train themselves to use the right words in the right context. Indeed, Acura’s navigation system, with voice recognition derived from I.B.M.’s ViaVoice software, can even perform concierge-in-a-car tricks like finding an address based on a phone number dictated by the driver. “Voice destination entry, or V.D.E., has been a big thing these days,” John Watts, manager for product planning at Honda’s Acura division, said. Today, navigation systems in luxury cars like the Acura MDX can respond to perfunctory commands, like “Find the nearest gas station.” Large vocabularies need to be stored on hard disk drives and processors have to work with complex language models - the combination of linguistics and statistics - to interpret what is being said. In part, speech recognition in cars has been held back by the limited amount of onboard computer memory and the processing demands of the software.
But so far automobiles have been limited to basic commands.
have made it possible for a PC to take dictation.
Already, desktop software like Dragon NaturallySpeaking from Nuance Communications and ViaVoice from I.B.M. This HAL 9000 fantasy is behind the development of voice recognition programs that enable people to converse with (or at least have their spoken commands understood by) machines. Voice recognition is, in theory at least, a relief from the human-machine interfaces that most drivers can only rate as frustrating. It wasn’t long before revelation struck: what we really wanted was not cars that talked, but listened.ĭecades later, voice recognition systems are a staple of the luxury class, an alternative to buttons, joysticks and touch-screens that control onboard entertainment systems and satellite navigation devices. NOT long after computers settled into cars to impose their electronic authority on engines, transmissions and safety systems, microprocessors began speaking up, nagging passengers to buckle the seat belts and issuing alerts about doors that were ajar.