Was really impressed with the Hyundai Genesis!

[Fight4Hemp]

That's ridiculous - speech recognition tech/algorithms have improved significantly over time. The fact is that Hyundai appears to have put very little effort into it and it shows: one has to look no further than the built in voice recognition that ABOLUTELY SUCKS versus the Google search voice recognition which is pretty flawless - both using the same microphone (speakers are not used to pick up sound, btw).

Thanks for that correction lol, yeah it has improved but they are still using the same technology they used 30 years ago according to one show I watched about 3 years ago on discovery channel. I don't claim I studied the matter intensely(obviously by my speaker statement .) the overall impression I got with the show was how far behind voice recognition is at its core. Not at the great things companies are doing with software or (microphones) but the basic principle of their method. Also when it comes to safety the university of Utah in 2014 discovered that you are 4 times more distracted while using voice recognition software. Really though if you can't handle driving and maybe typing a phone number you probably don't need to be driving in the first place. It's always best to pull over if you have the reflexes of a turtle. Knowing your limits is key to staying safe.

 

[Nikgen]

Thanks for that correction lol, yeah it has improved but they are still using the same technology they used 30 years ago according to one show I watched about 3 years ago on discovery channel. I don't claim I studied the matter intensely(obviously by my speaker statement .) the overall impression I got with the show was how far behind voice recognition is at its core. Not at the great things companies are doing with software or (microphones) but the basic principle of their method. Also when it comes to safety the university of Utah in 2014 discovered that you are 4 times more distracted while using voice recognition software. Really though if you can't handle driving and maybe typing a phone number you probably don't need to be driving in the first place. It's always best to pull over if you have the reflexes of a turtle. Knowing your limits is key to staying safe.

Geez - you seem to derive much of your knowledge on topics you post about from TV and You Tube. Voice recognition technology is all about the software (assuming the mic is of required minimum quality to record the voice with sufficient clarity) - i'm not buying the results of that study!

And sorry, good voice recognition is far less distracting than physically typing on your phone or car Nav screen.

 

[MoabGen]

You're lucky - I rarely ever got things even remotely close to what I asked for, so I have not used it in over a year. The Bluelink Google Search, on the other hand, pretty much gets it right everytime (as does Waze on my phone).

Hyundai was just flat out lazy in both voice recognition quality and breadth of vehicle systems integration with voice control. A pretty big oversight in an otherwise really well thought out vehicle.

Maybe that was their goal, water down the VR in hopes people would spend $100/yr to pay for the Google search

 

[Nikgen]

Maybe that was their goal, water down the VR in hopes people would spend $100/yr to pay for the Google search

You never know these days

 

[daviderle]

That's ridiculous - speech recognition tech/algorithms have improved significantly over time. The fact is that Hyundai appears to have put very little effort into it and it shows: one has to look no further than the built in voice recognition that ABOLUTELY SUCKS versus the Google search voice recognition which is pretty flawless - both using the same microphone (speakers are not used to pick up sound, btw).

I will let you in on a little secret, Hyundai has nothing to do with the speech engine selection / technology, that decision is done by Lexicon, the system used by them is based on 'Nuance auto', an offshoot of Nuance v5 Windows version 2009-2014 and heavily modified v6 Linux version on some 2015 trims (in 2014 Nuance is about the only speech recognition engine left standing that can be installed on an end device, outside the big boys left with their cloud based offerings)

To give you an idea how far behind the 'auto' version is, currently when you call your favorite institution and hear "please speak or enter your account number...", these systems are most likely running on at least version 10 or above.

In addition speech recognition takes lots of processing power, for example adaptive 'say anything' technology, these days run on duel Xeon, 12 core running at 3.4Ghz for a typical implementation dedicated to speech processing, while the 'tiny' 800mhz-1gh processor found in our head units, can only do a very limited recognition tasks based on a very limited and controlled grammar set.

Google. Apple and now Amazon have dedicated data centers, filed with thousands clustered servers running AI based software, that can actually learn your particular voice print which makes it preform better over time based on millions of samples.

Hyundai or anyone else for that matter can not compete at that level in this arena, unfortunately the big boys do not have an external API for their systems, Amazon is trying with their AWS "Echo" but they are still in the very early stages of this effort and mainly focused on home automation.

So apple and Android auto are the best next thing, unfortunately it is a catch twenty two, where the auto industry does not want to be reliant on the presence of the Internet for the car to function on a daily bases. [EDIT] or being forced to charge for this service to play nicely with Google and/or Apple for this enhanced functionality, so they continue to use 15 year old obsolete technology...

 

[jsmit86]

Voice recognition can be a bit inconsistent on any car.

I can't recall the command I had tried on my 2011 Sonata, but the response was "Find Korean Restaurant?" That cracked me up. While I can't recall the command, I tried it several times, and the car kept asking me if I wanted to... "Find Korean restaurant?"

We also own a Dodge Journey... I wonder now the Chrysler is owned by Fiat.... would I get "Find Italian Restaurant?"

 

[Fight4Hemp]

Geez - you seem to derive much of your knowledge on topics you post about from TV and You Tube. Voice recognition technology is all about the software (assuming the mic is of required minimum quality to record the voice with sufficient clarity) - i'm not buying the results of that study!

And sorry, good voice recognition is far less distracting than physically typing on your phone or car Nav screen.

You can dismiss it anyway you like. It's not up to me or do I even care if you choose to accept a study conducted by an university. As always nik I can always count on you to disagree and correct which doesn't bother me. I'm not going to beat a dead horse but my main point of all this is try talking with a lisp or accent next time you are trying to look something up using VR and then tell me how great the software is. Remember just because you can run your house on dc doesn't mean there isn't a better method.

 

[Nikgen]

You can dismiss it anyway you like. It's not up to me or do I even care if you choose to accept a study conducted by an university. As always nik I can always count on you to disagree and correct which doesn't bother me. I'm not going to beat a dead horse but my main point of all this is try talking with a lisp or accent next time you are trying to look something up using VR and then tell me how great the software is. Remember just because you can run your house on dc doesn't mean there isn't a better method.

My point was that there is no other way to handle voice recognition. At it's core, the basic technology involves optimized software algorithms and, as Daviderle pointed out, sufficient processing power to run that software. Contrary to your suggestion that no progress has been made and the technology is the same as 30 years ago, there has been HUGE progress made with this technology over time.

Obviously, an accent or a lisp would pose a problem for any system - but i have neither, so I would expect the Hyundai (or Lexicon, I guess) system to be reasonably accurate - it isn't! In fact, it's the worst voice recognition system I've ever used. This makes it effectively useless in many cases as it would be faster to just type it in yourself. Thankfully, we have Google Search and Siri Eyes free - both have excellent voice recognition.

University studies are not always correct or maybe you misinterpreted the type of VR usage they were studying.

For the most common task of making a phone call, I just don't see how pressing a button on my steering wheel and saying "call home" without ever having to look away from the road can possibly be more distracting than pulling up the phone app on a phone or in the car system and searching through contacts or typing in the phone number - all of which require looking away from the road.

The same is true for Nav. How can pressing a button on a mirror, saying an address, confirming with a button press and starting the route with one more press be more distracting than opening the NAV app with a few button presses, typing an address in manually and still having the additional button presses to confirm and start the route?

Common sense says the study you reference was flawed or looking at something other than what you suggested.
.

 

[daviderle]

In addition to hardware and software limitation present in our cars, there is another technical reason as to why the recognition performance of our units can be considered to be sub par, as the current recognition software can only process narrow band audio (think old analogue lines with 300 to 3.4khz frequency range ), while both Siri and Google use super wide audio (50 - 7khz range) which also account for their accuracy, as these services are able to do a better job of extracting the actual human 'speech' for any given audio stream...

 

[Nikgen]

In addition to hardware and software limitation present in our cars, there is another technical reason as to why the recognition performance of our units can be considered to be sub par, as the current recognition software can only process narrow band audio (think old analogue lines with 300 to 3.4khz frequency range ), while both Siri and Google use super wide audio (50 - 7khz range) which also account for their accuracy, as these services are able to do a better job of extracting the actual human 'speech' for any given audio stream...

Thanks - you have this stuff down . What I still don't understand is why the built in Hyundai system is so bad - it gets very basic stuff wrong all the time that I've not had probkems with before in other cars.

______________________________

Help support this site so it can continue supporting you!

 

[daviderle]

Thanks - you have this stuff down . What I still don't understand is why the built in Hyundai system is so bad - it gets very basic stuff wrong all the time that I've not had probkems with before in other cars.

Thank you, let us just say, electronics and performance are my passion and I love to take things apart and put them back together, alas, my commitments do not allow me enough time to dedicate to these endeavors, that is why you may have noticed sometimes I drop off the map for couple months...

Back to the question at hand, the system in our car is grXML driven with very specific grammar for each menu, which means, you must be absolutely specific for each menu, for example in the 2009 4.6 I have to say "go home", while in the 2013 R-Spec I have to say "navigate home" and if I say anything else in either car, it will get it wrong! "Call home" for example!

Now there is another thing that might be useful to know, for the 2013-14 and the 2015 models, they enabled background noise adaptation, which means the system will try extrapolate the background noise for better voice recognition. To utilize this feature you need press the voice command button, wait for the beep, then wait for a full two seconds before speaking your command. In general this new feature does provide a significant recognition improvement.

Second suggestion, as I recall there is a quick reference card that comes with our vehicles and I found it essential to memorize this card which illustrates the exact commands for each menu to get the most out of the speech recognition feature provided by each individual model.

 

[Nikgen]

Thank you, let us just say, electronics and performance are my passion and I love to take things apart and put them back together, alas, my commitments do not allow me enough time to dedicate to these endeavors, that is why you may have noticed sometimes I drop off the map for couple months...

Back to the question at hand, the system in our car is grXML driven with very specific grammar for each menu, which means, you must be absolutely specific for each menu, for example in the 2009 4.6 I have to say "go home", while in the 2013 R-Spec I have to say "navigate home" and if I say anything else in either car, it will get it wrong! "Call home" for example!

Now there is another thing that might be useful to know, for the 2013-14 and the 2015 models, they enabled background noise adaptation, which means the system will try extrapolate the background noise for better voice recognition. To utilize this feature you need press the voice command button, wait for the beep, then wait for a full two seconds before speaking your command. In general this new feature does provide a significant recognition improvement.

Second suggestion, as I recall there is a quick reference card that comes with our vehicles and I found it essential to memorize this card that illustrates the exact commands for each menu to get the most out of the speech recognition feature provided by each individual model.

Thanks - I'll try that. I do get the need to use the exact commands - the bigger issue seems to be when you dictate an address, many times the results that come back are not even close to what was said. Not a biggie anymore as I just use Google Search, but disappointing.

 

[daviderle]

Thanks - I'll try that. I do get the need to use the exact commands - the bigger issue seems to be when you dictate an address, many times the results that come back are not even close to what was said. Not a biggie anymore as I just use Google Search, but disappointing.

"Address capture" as it is known, is pretty difficult to preform locally on the tiny processor without the access to a more powerful backend system due to large number of data set permutations.

However, another suggestion is to speak the address slowly and as much as possible in a uniform tone/loudness, (just as if you are dictating the address to some foreign national in a data center located somewhere in Asia).

 

[SpoofeeDotCom]

05? Are you serious? That was 11 years ago.

 

[Fight4Hemp]

Obviously, an accent or a lisp would pose a problem for any system - but i have neither,
.

Really? I always pretend that you have a Scottish accent and a lisp every time I read your comments.

 

[Midnightsun]

Well my 2010 and 2012 never missed a beat when I said "call home" My 2015 works 30% of the time and sometime I have to say it 3 times to get it. A couple of times I gave up, pulled out the iPhone and hit the speed dial on it. For what ever reason when I say call home I repeatedly get a name in my address book "Chantal Baliff". Yep it sucks to the 10th degree.

Question to you out there. Do you have issues when you say call home or is my unit in need of replacement?

I also have a Ford F150 with sync, rock solid, actually perfect if you ask me.

 

[ksoze]

Now there is another thing that might be useful to know, for the 2013-14 and the 2015 models, they enabled background noise adaptation, which means the system will try extrapolate the background noise for better voice recognition. To utilize this feature you need press the voice command button, wait for the beep, then wait for a full two seconds before speaking your command. In general this new feature does provide a significant recognition improvement.

In the 2015, holding the voice button for a second or two will bring up Siri.

 

[jime]

I had a lot of trouble with the voice command - until I found that the mic volume was too high. Try adjusting the volume so that the power bar at the top of the display just reaches the maximum level when you are speaking.

 

[Mark_888]

Speech recognition has not advanced in a very long time, it is really old technology and highly outdated. True they have innovated uses as in voice to text or hands free commands but the same old sorry technology is still used to determine what you said. They also improved the speaker to pick up your voice and clarify what came out your mouth but it's the same cake just with different frosting.

How would you know? 11 years ago, you were only 15 years old.

Speech recognition in cars has improved significantly in the last 11 years, not to mention that it is much more prevalent now than in 2005.

 

[Fight4Hemp]

How would you know? 11 years ago, you were only 15 years old.

Speech recognition in cars has improved significantly in the last 11 years, not to mention that it is much more prevalent now than in 2005.

Again I don't know why I need to explain this again im just explaining what i seen on tv. I didn't do any research Jesus Christ that's like saying how do you know who the king of England was during the 30 year war in 1618 you wasn't there...get real