Design Article
TECH TRENDS: Front-end voice activation improves auto safety
Todd Mozer, Sensory
10/20/2011 3:00 AM EDT
Why buttons for so long at Stage 1?
There are several reasons speech recognition has required buttons to activate rather than voice activation. The main reason has been that buttons, although distracting and expensive, are quite reliable and responsive, even in noisy environments. The car is a challenging environment for speech recognizers and a voice activation word must respond with windows down, radios on, and without the aid of having the microphone a few inches from the speakers’ mouth. Traditional speech technologies are reliable at responding in quiet environments but not the high noise of a car.
The requirement of a speedy response time further complicates this challenge. Speech recognizers often require hundreds of milliseconds just to determine the user is done talking before starting to process the speech. This time delay might be acceptable from a recognition system to yield an answer or reply to the consumer. However, at Stage 1, the response of the activation is calling up another more sophisticated recognizer at Stage 2, and consumers will not accept a delay lasting much more than the time it takes to press a button. The longer the delay, the more likely a recognition failure occurs at Stage 2 because users might start talking before the Stage 2 recognizer is ready to listen.
What’s new today?
Recent advances by Sensory with the TrulyHandsfree Voice Control technology make buttons unnecessary. The TrulyHandsfree solution is robust to noise, small in footprint and power consumption, accurate in responding when spoken to, and fast in response time. Since TrulyHandsfree’s recent introduction it is already appearing on the market in devices from toys to telephones.
Stage 1 implementation
Several major hurdles were overcome in creating the Stage 1 recognition system. A new probabilistic-based model allows the recognition to work effectively even with the radio on, windows down, and motor noise. The nature of the probabilistic model allows the recognizer to determine a spoken trigger phrase is being said in real time, because it doesn’t require 100% confidence to accept the phrase, so it can actually respond with 90% of the phrase spoken. This approach helps to improve performance in noise and with stronger accents.
Typically, the most difficult part of implementing a voice user interface is in the system specification and dialog design. For existing speech based systems with button based activation, it is a simple procedure to advance to a voice activated mode by adding a Stage 1 recognizer. The simplicity of this “incremental” change is accomplished because the user experience is enhanced without the need to redefine it. Existing dialog designs can be used with the enhancement of replacing the button press. Usually this transition can be done with a small amount of added software taking only a few hundred kilobytes of memory, and around 40 MIPS of processing power.
Inexpensive dedicated speech chips (such as Sensory’s NLP-5X) are also available making the addition of voice activation or simple command and control functions simple, fast, flexible, and relatively inexpensive.
What the future holds
In car speech systems will see substantial improvements over the coming years. The automotive voice user interface, always desirable in the car but only recently usable, will be optimized for ease of use and become quite essential for convenience, productivity and safety. The front end voice activation is a great new step in this direction and advancements in noise-robust technology and deciphering “meaning” will continue to fuel the utility of speech technologies in car.
Todd Mozer is CEO of Sensory. He can be reached at tmozer@sensoryinc.com.
There are several reasons speech recognition has required buttons to activate rather than voice activation. The main reason has been that buttons, although distracting and expensive, are quite reliable and responsive, even in noisy environments. The car is a challenging environment for speech recognizers and a voice activation word must respond with windows down, radios on, and without the aid of having the microphone a few inches from the speakers’ mouth. Traditional speech technologies are reliable at responding in quiet environments but not the high noise of a car.
The requirement of a speedy response time further complicates this challenge. Speech recognizers often require hundreds of milliseconds just to determine the user is done talking before starting to process the speech. This time delay might be acceptable from a recognition system to yield an answer or reply to the consumer. However, at Stage 1, the response of the activation is calling up another more sophisticated recognizer at Stage 2, and consumers will not accept a delay lasting much more than the time it takes to press a button. The longer the delay, the more likely a recognition failure occurs at Stage 2 because users might start talking before the Stage 2 recognizer is ready to listen.
What’s new today?
Recent advances by Sensory with the TrulyHandsfree Voice Control technology make buttons unnecessary. The TrulyHandsfree solution is robust to noise, small in footprint and power consumption, accurate in responding when spoken to, and fast in response time. Since TrulyHandsfree’s recent introduction it is already appearing on the market in devices from toys to telephones.
Stage 1 implementation
Several major hurdles were overcome in creating the Stage 1 recognition system. A new probabilistic-based model allows the recognition to work effectively even with the radio on, windows down, and motor noise. The nature of the probabilistic model allows the recognizer to determine a spoken trigger phrase is being said in real time, because it doesn’t require 100% confidence to accept the phrase, so it can actually respond with 90% of the phrase spoken. This approach helps to improve performance in noise and with stronger accents.
Typically, the most difficult part of implementing a voice user interface is in the system specification and dialog design. For existing speech based systems with button based activation, it is a simple procedure to advance to a voice activated mode by adding a Stage 1 recognizer. The simplicity of this “incremental” change is accomplished because the user experience is enhanced without the need to redefine it. Existing dialog designs can be used with the enhancement of replacing the button press. Usually this transition can be done with a small amount of added software taking only a few hundred kilobytes of memory, and around 40 MIPS of processing power.
Inexpensive dedicated speech chips (such as Sensory’s NLP-5X) are also available making the addition of voice activation or simple command and control functions simple, fast, flexible, and relatively inexpensive.
What the future holds
In car speech systems will see substantial improvements over the coming years. The automotive voice user interface, always desirable in the car but only recently usable, will be optimized for ease of use and become quite essential for convenience, productivity and safety. The front end voice activation is a great new step in this direction and advancements in noise-robust technology and deciphering “meaning” will continue to fuel the utility of speech technologies in car.
Todd Mozer is CEO of Sensory. He can be reached at tmozer@sensoryinc.com.
_________________________________
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
Dr DSP
10/20/2011 5:24 PM EDT
Just what we need. Let's put lots of capabilities into the car to distract us and then more to take up the slack when we are distracted. Let the car do the driving and then we can just watch a movie on the wind shield.
Sign in to Reply
ndancer
10/21/2011 4:22 PM EDT
I have, on a 2010 Prius, a voice activated GPS / radio module. On a recent trip, I tried, "Zoom out", a legal command, by the way, and got everything from "Show restaurants" to "CD on". And we're going to trust these things to drive the car?
Guys, this is not a good thing.
It could be greatly improved if it asked, "I heard CD on. Is this correct?" That alone would decrease frustration level.
I no longer even try to use voice commands, although I was raised in Colorado, and, according to most people, enunciate clearly enough, but not, apparently, clearly enough for slockware, er, I'm sorry, software.
Sign in to Reply
Bert22306
10/22/2011 6:35 PM EDT
Maybe eventually, but I'm skeptical still. I have an OnStar system that is voice activated, but it's not fool proof. Oddly enough, it has most problems when I tell it to call "home." "Home" is one word it finds baffling. Needs repeating just about every time. So, just imagine being stressed out, in some critical situation, and barking out a command the system can't decipher. Perhaps because your voice sounds different when stressed. Come now.
On the subject of hands-free calling, I've read more than one report that says it's just about as distracting as using a cell phone. While I don't dispute that, I'm curious why talking hands-free on a phone should be any different from having a conversation in the car, with a passenger, while driving.
All I can come up with is that while driving and conversing with a passenger, the driver feels freer to stop talking when the driving requires more attention? And on the phone, you can't just leave the other guy hanging? Or perhaps conversing over a low-fidelity audio link, like voice telephone links, is simply more of a burden on the brain, than in person conversations? I mean, it's not like the driver can actually look at the person he's speaking to, while driving, so the missing visual cues can't be the difference in this phone while driving matter.
Sign in to Reply
prabhakar_deosthali
10/24/2011 1:31 AM EDT
I also have serious doubts about the efficacy of such voice activated systems in critical situations when your voice may not remain the same with which you would have trained your voice system.
It is better to have some touch pad kind of systems with one touch command facility , on the dashboard.
Sign in to Reply
ndancer
10/28/2011 12:16 PM EDT
How about have the car automatically detect, A: when traffic is too heavy to be safe, or, B: when you're driving like a jerk, and then, auto-magically, turn off the radio, TV, GPS, telephone, ipod, and everything else that could possibly be distracting.
Might cut down on the DWS syndrome (Driving While Stupid).
Sign in to Reply
WKetel
10/28/2011 1:16 PM EDT
I agree that voice commands are very complex to handle, and the comment about "probablistic algorithms means that the system would always assume the most common command. One option that could help might be similar to what we saw in Star Trek, where Kirk would first say "computer", and then state his request.
Of course, all of this technology will certainly serve to provide more distractions and reduce safety, since most human minds can't handle large amounts of distraction without ignoring the more important task of driving. This is not just some wild assertion, it has been proved with research several times.
Sign in to Reply
eckna
2/18/2013 5:59 AM EST
jkhh
Sign in to Reply