Design Article
Comment
VUI Guy
The CPU hit factor is not something to be taken lightly, which is why companies ...
rajuchaluva
Speech recognition cuts driver distraction
Rick DeMeis
8/5/2011 10:57 AM EDT
Sidebars: Experience, history, and off into the cloud
Early on the road
This editor had the chance to sample early automotive voice recognition in driving one of the first Jaguar S-Types in the late '90s. A set of simple, structured commands was used to control the audio/radio system and climate control. I remember when trying to "Set temperature, 72 degrees" the system would not understand me, but I could set it to 70 or 74 (which was recognized), and then use an "up" or "down" command to bring it to 72.
How y'all doin'?
Nuance's Brian Radloff says the company's speech recognition systems come in 28 languages for automotive applications. Within these language models, a variety of accents are built-in, such as U.S. regional dialects. Also algorithms in the recognition engine adapt to the speaker by learning from which system responses are corrected by the user. In contrast, some early systems were "trained" to the user's voice by listening to a series of test words—which would probably take a long time with today's more extensive vocabularies (grammars).
Hyundai goes off into the cloud (with handholding)
Hyundai is currently introducing its Blue Link telematics system with voice recognition features that are partially based off-board the vehicle. The system was developed by Hyundai and telematics services provider ATX Group, and uses Nuance and Vlingo speech recognition software.
The system marries speech recognition with agent assist, says ATX' Don Tryon, senior director, account management. Initial interactive voice recognition at a conversational pace is done with Nuance software. The response to the request, say for directions or a point-of-interest, is analyzed and monitored by Vlingo tools (which have a large cloud-based library of "utterances"). If the system has a confidence level on the order of 90% surity of the command, functionality remains on the vehicle. If it falls below, an agent—using a list of options provided by Vlingo—will talk to the driver for clarification and assistance.
In developing Blue Link, the importance of development kits from the software providers cannot be underestimated, states Michael Deitz, senior group manager for Connected Car at Hyundai. "The sooner a system can be in the car for testing the better," he adds. The kits enable early checking of system performance in the actual vehicle's acoustic environment.
An embarrassment of recognition riches
When we talk about challenges in voice recognition, as engineers we tend to think in terms of signal-to-noise ratios and algorithms. But QNX' Scott Pennock notes another hurdle is simply making users aware of speech-enabled features and functions—so they can have a more pleasant and productive drive.
That thought is echoed by Ford's Brigitte Richardson. "People get into a car with so many features, such as traffic information, phones, and satellite and HD radio, but they don't know what you can control by voice. Likewise, they won't know what to say to give a command and interact with the car," she notes. "People are reluctant to go through a system-given tutorial or read a large manual (as we provided in 2009). Many users are unaware we provide weather radar or 'one-shot' destination entry (just by saying an address), which are very useful."
Because of such tendencies, Ford and Nuance added much more grammar with lots of aliases (synonyms that are embedded but not "advertised") to make its SYNCTM system easier to access with minimal "training," Richardson says. She adds that the overall "take rate" for SYNC since its introduction in 2007, thanks to its features and utility, has been 79% across the Ford brand. The technology is complex, Richardson states, but that complexity is in the voice-recognition engine—simplifying the user interface is key to acceptance and usability.
This article was originally published in EDN.
Early on the road
This editor had the chance to sample early automotive voice recognition in driving one of the first Jaguar S-Types in the late '90s. A set of simple, structured commands was used to control the audio/radio system and climate control. I remember when trying to "Set temperature, 72 degrees" the system would not understand me, but I could set it to 70 or 74 (which was recognized), and then use an "up" or "down" command to bring it to 72.
How y'all doin'?
Nuance's Brian Radloff says the company's speech recognition systems come in 28 languages for automotive applications. Within these language models, a variety of accents are built-in, such as U.S. regional dialects. Also algorithms in the recognition engine adapt to the speaker by learning from which system responses are corrected by the user. In contrast, some early systems were "trained" to the user's voice by listening to a series of test words—which would probably take a long time with today's more extensive vocabularies (grammars).
Hyundai goes off into the cloud (with handholding)
Hyundai is currently introducing its Blue Link telematics system with voice recognition features that are partially based off-board the vehicle. The system was developed by Hyundai and telematics services provider ATX Group, and uses Nuance and Vlingo speech recognition software.
With voice recognition, the physical interface for Hyundai's Blue Link telematics system is simple and clean.
The system marries speech recognition with agent assist, says ATX' Don Tryon, senior director, account management. Initial interactive voice recognition at a conversational pace is done with Nuance software. The response to the request, say for directions or a point-of-interest, is analyzed and monitored by Vlingo tools (which have a large cloud-based library of "utterances"). If the system has a confidence level on the order of 90% surity of the command, functionality remains on the vehicle. If it falls below, an agent—using a list of options provided by Vlingo—will talk to the driver for clarification and assistance.
In developing Blue Link, the importance of development kits from the software providers cannot be underestimated, states Michael Deitz, senior group manager for Connected Car at Hyundai. "The sooner a system can be in the car for testing the better," he adds. The kits enable early checking of system performance in the actual vehicle's acoustic environment.
An embarrassment of recognition riches
When we talk about challenges in voice recognition, as engineers we tend to think in terms of signal-to-noise ratios and algorithms. But QNX' Scott Pennock notes another hurdle is simply making users aware of speech-enabled features and functions—so they can have a more pleasant and productive drive.
That thought is echoed by Ford's Brigitte Richardson. "People get into a car with so many features, such as traffic information, phones, and satellite and HD radio, but they don't know what you can control by voice. Likewise, they won't know what to say to give a command and interact with the car," she notes. "People are reluctant to go through a system-given tutorial or read a large manual (as we provided in 2009). Many users are unaware we provide weather radar or 'one-shot' destination entry (just by saying an address), which are very useful."
Ford's SYNC voice recognition has evolved rapidly since being introduced in 2007.
Because of such tendencies, Ford and Nuance added much more grammar with lots of aliases (synonyms that are embedded but not "advertised") to make its SYNCTM system easier to access with minimal "training," Richardson says. She adds that the overall "take rate" for SYNC since its introduction in 2007, thanks to its features and utility, has been 79% across the Ford brand. The technology is complex, Richardson states, but that complexity is in the voice-recognition engine—simplifying the user interface is key to acceptance and usability.
This article was originally published in EDN.
____________________________
If you liked this article, go to the Automotive Designline home page for the latest in automotive electronics design, technology, trends, products, and news. Also, get a weekly highlights update delivered directly to your inbox by signing up for our weekly automotive electronics newsletter here.
|
|
|
|
|
Navigate to related information
agk
8/6/2011 7:04 AM EDT
This is a nice system. The driver can keep the hands on the steering wheel and eyes on the road while operating all the gadjets by giving voice commands. The driver can control the entertainment gadgets,navigation,get the vehicle information,climate control and have telephonic conversation by voice commands.
Sign in to Reply
cdhmanning
8/10/2011 1:39 AM EDT
It is not just keeping your hands on the wheel and eyes on the road that is important. You need to keep your brain engaged too.
If you are swearing at the machine because it is not hearing you properly then you are probably not in the right mood for driving!
It has been noted with Bluetooth headsets (which are legal in many areas) that these are as distracting as holding a cellphone to your ear during a voice call.
I must say I have not used voice recognition for a long time, but anyone that remembers the Microsoft VR fiasco will remember how bad it can be!
I once worked for a telecom company where we tested some VR gear that had been trained to understand British voices. None of the British guys in the office could make it work. Nor me (South African accent). The only guy who could make it work was a Fijian Indian guy!
Sign in to Reply
rajuchaluva
8/25/2011 2:54 AM EDT
Thats true!
Great work
Sign in to Reply
prabhakar_deosthali
8/8/2011 2:59 AM EDT
Sure! Voice recognition UIs can definitely reduce the distractions for the driver. But some of those voice conversation themselves may be too distracting, for example if the driver happens to get a nagging call from his wife! There will be such situations when the drivers eyes will be on the road, his hands on the steering wheel but his mind getting dragged into some other world and a driver in everybody knows well what disasters can happen when your mind is not on the job at hand.
Sign in to Reply
hm
8/8/2011 5:47 AM EDT
Very good progress but it has long way to go. Soon CMOS Camera and advance image processing will add to new UI.
Sign in to Reply
Rick DeMeis
8/16/2011 1:58 PM EDT
I had the chance recently to interact the SYNC system in a Ford Edge. It was easy to get to use, with responses to most of the intuitive commands I could think up to control the audio and the climate control. Unfortunately, this car did not have a navigation system, which is where I think the utility of speech recognition can prove most useful.
I hope to use some other Nuance-based voice recognition products in the near future, and will report on the results.
Sign in to Reply
selinz
8/24/2011 2:47 PM EDT
I like the system and guess what, my phone has the same function. I just hope we can keep the legislators from limiting the usability.
Sign in to Reply
VUI Guy
6/21/2012 4:47 AM EDT
The CPU hit factor is not something to be taken lightly, which is why companies turn to a vendor like Rubidium, which offers a small footprint, low resource, cost effective solution as opposed to the larger Nuance product.
Sign in to Reply