CALIFORNIA, USA – Cloud-based voice recognition (VR) on consumer electronics devices didn’t start with Siri. But in the same way that Apple has helped refine other technology interactions that were first to market, Siri now personifies voice recognition for many people. And it’s become the poster child for off-board VR – whether most users realize they’re talking to the cloud or not.
Considering all the tech that now comes in cars, drivers desperately need a Siri riding shotgun. Car companies often play the VR card when grilled by government officials and safety advocates on how drivers are supposed to control all the latest dashboard gadgets while keeping their eyes on the road and hands on the wheel.
But anyone who has tried most VR systems in cars will tell you that they rank somewhere below, say, Sri Lankan tech support in terms of two-way verbal comprehension. While VR has incrementally improved, it’s still the tech feature that’s most irksome to car owners. “One of the biggest elements that pop up on J.D. Power & Associates’Initial Quality Study is voice recognition errors,” said Mark C. Boyadjis, an analyst who covers automotive for IHS Automotive. “There’s a huge push right now to make voice recognition better.”
Nuance, the company behind the popular Dragon suite of speech-to-text software, is the VR provider of choice for automotive and, in fact, dominates the market. “It’s not really a question,” Boyadjis added. “They have the majority worldwide of all automotive voice recognition systems.” As Nuance goes, so goes automotive VR.
So when Nuance announced this week that it’s moving in-car VR to the cloud with its new Dragon Drive! platform, it signaled a major shift in the technology. And that your next car may actually understand what you’re saying.
Dragon Drive! is designed to make accessing connected-car functions – from finding the cheapest fuel nearby to keeping up with your Facebook feed – safe and more convenient. And allows drivers do it through more conversational interaction, meaning you won’t have to speak like a robot when giving the car commands.
Ed Chrumka, Nuance’s senior product manager for connected car services, told Wired that Dragon Drive! is essentially an automotive-grade version of the company’s Dragon Go! smartphone application. “We’re taking that same capability and extending it to inside the vehicle,” he said. But Dragon Drive! isn’t completely off-board, and no automotive VR engine probably ever will be. And it’s not even the first off-board VR application.
There’s already a blending of on-board and off-board VR in some cars, noted Boyadjis. “Ford Sync has onboard for Bluetooth, audio and navigation, but when you access Sync Services such as Traffic Directions and Information, you‘re talking to off-board voice recognition.” Chrumka points out that there’s an important distinction. “With Sync Services, everything goes over a voice channel. Dragon Drive! will go over a data channel so we can capture higher bandwidth, which leads to better accuracy.”
Jim Buczkowski, director of electronics research for Ford, points out that although Sync Services uses a voice channel, the feature works with any dumb phone via Bluetooth and doesn’t require a smartphone with a data plan. “But, certainly, there’s some advantage to [using a data channel] and that will continue as smartphone use grows.”
Chrumka added that Dragon Drive! works with either a smartphone-tethering approach like Ford’s Sync App Link feature or an embedded cellular modem a la OnStar. But even as smartphones become prevalent and more automakers take a “hybrid” approach to connectivity – using an onboard modem for critical services like crash notification and the driver’s connected smartphone to, say, stream Pandora – cloud-based VR will likely be tied into the car’s onboard system for some time.
One reason is that the car environment presents unique challenges smartphone VR applications don’t have to face, explains Brian Radloff, director of automotive solutions for Nuance. “On a handset with a close-talk microphone, that’s clearly different than when you’re driving in a vehicle with a far-talk microphone, and you have cabin noise and other noise that gets picked up,” he said. “We’ve designed very specific acoustic models … so that the system is robust to noise. We’ve also implemented software that allows us to de-noise the sound signal by focusing an acoustic beam toward the speaker.”
Another automotive obstacle – and the reason we haven’t seen cloud-based VR in the car until now – is the long product development cycles in automotive. “It can take 18, 24, sometimes even 36 months to develop a specification, and then it has to go through full-vehicle validation,” Radloff said.
Nuance won’t disclose which vehicle we’ll see Dragon Drive! in first, but would only say that it will be included in cars later this year. “The initial use case is dictating a message, and then that can be formatted into an SMS message, an e-mail or even for things like Twitter and Facebook updates,” Chrumka said.
Moving VR to the cloud will also allow the car to better understand casual speech like, “I want to hear some Hendrix.” “Behind the cloud is a lot more analysis and memory capability, so the vocabulary can be much larger,” Buczkowski explained. “The ability to [convert] a set of utterances into a meaningful sentence or question is easier leveraging cloud services than trying to do it onboard.”
“The amount of things you can do with VR really depends upon off-board connectivity,” Boyadjis added, “because there’s no way to get that kind of processing power in the car. If you’re going to have very contextual elements where you can just mumble something and it’s going to pick up key words and analyze it to correctly establish what you’re saying, there has to be hybrid voice recognition. I don’t think we’ll ever go back to fully static on-board VR systems.”
Buzkowski pointed out one potential downside to cloud-based VR: “You’re always going to have the possibility for latency,” he said. “You have to make that roundtrip out to the processing in the cloud. And it makes no sense to go out to cloud to change the radio station or climate temperature or anything that’s local to the vehicle.”
And one issue with off-board VR that should be obvious to anyone on AT&T’s network: You can’t ask the cloud to find the nearest Starbucks if you don’t have a good connection.
Source: WIRED
