Announcement

Collapse
No announcement yet.

How does Honda do voice rec without training?

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • How does Honda do voice rec without training?

    I've gone through my pc and done the voice training a BUNCH and now it recognizes most everything I say, but it doesn't take any commands from strangers (not that that's necessarily bad). My friend has a new Honda Accord and the voice rec in it is phenominal. I sat in it for the first time and was able to voice control the sound system and nav system without ANY training at all. Is there a way to do this on a pc?
    Epia-M II M10000 Nehemiah WinXP sp2 Travla C134 case Carnetix 1260 power Xenarc 700TS XM Direct Toslink Extigy USB 5.1 Garmin GPS18 Linksys 802.11g wireless Nokia 6230 Bluetooth Memphis Amps Focal Speakers Frontend: RoadRunner

  • #2
    some voice recognition algorithms are better than others.

    I had a SE T616 and before that a Moto v70. Now, I have a Moto V600 and the voice recognition on the SE was 100x better than any of those Motorolas.

    Comment


    • #3
      The Voiceflux software I was using didn't require any learning and it seems to work quite well. I'm not sure about how much better or worse the recognition is compared to a trained program though.

      I guess it also depends on what you need the software to do.

      Cheers,

      Kris
      How Much Horsepower Can I have And Still Go To Heaven?

      Comment


      • #4
        Ok.... what is VoiceFlux?

        But yeh, why would you want the computer taking commands from strangers? My firends would go nuts when they are in the car because they would be able to control the computer
        PowerVoice v1 | NaviVoice Source
        GammaControl v2.4
        SKINbedder v3

        1995 Lexus SC300

        Comment


        • #5
          Honda uses embedded viavoice running on QNX RTOS... I know two people that just got '05 Accord coupes and that stuff is killer. It's not without its problems though, and I did see it fail once or twice when they tried to show off (of course, the worst possible time )

          It's still pretty nice though... makes me wanna go look into it if it wasn't for the licensing fees.

          Comment


          • #6
            Originally posted by viguera
            Honda uses embedded viavoice running on QNX RTOS... I know two people that just got '05 Accord coupes and that stuff is killer. It's not without its problems though, and I did see it fail once or twice when they tried to show off (of course, the worst possible time )

            It's still pretty nice though... makes me wanna go look into it if it wasn't for the licensing fees.
            I think its best if we don't use a commercial product But being in the acura dealership did give me a idea: have a command to activate listening (so no saying "boris").... how does that sound?
            PowerVoice v1 | NaviVoice Source
            GammaControl v2.4
            SKINbedder v3

            1995 Lexus SC300

            Comment


            • #7
              I use the microsoft voice SDK and never trained it for command and control and get accuracy in the high 90s. The things that kill recognition is using a bad microphone, having lots of unfiltered ambient noise, or using a bad command set (i.e. 1000s of commands or similar sounding commands). If your getting bad recognition with your computer based voice recognition, more then likely its due to one of these things rather then the software not being as good. The reason why the ones in the honda are so good is because they can optimize the hardware and software together.
              StreetDeck.com Developer (I am Chuck)
              Get StreetDeck at http://www.streetdeck.com
              The Official StreetDeck Forums have moved, please visit us at http://www.streetdeck.com/forum for official support for Streetdeck.

              Comment


              • #8
                Originally posted by god_of_cpu
                I use the microsoft voice SDK and never trained it for command and control and get accuracy in the high 90s. The things that kill recognition is using a bad microphone, having lots of unfiltered ambient noise, or using a bad command set (i.e. 1000s of commands or similar sounding commands). If your getting bad recognition with your computer based voice recognition, more then likely its due to one of these things rather then the software not being as good.
                IMO I get excellent recognition. Some people do have accents, so they have to edit the commands to fit their accent.
                PowerVoice v1 | NaviVoice Source
                GammaControl v2.4
                SKINbedder v3

                1995 Lexus SC300

                Comment


                • #9
                  I'm not sure what the Honda stuff is used for, but i'd imagine that the reason it can understand almost anyone is because it has a ready defined command set, whereas the stuff we are using (well, i havent yet, but i'm planning to!) is used to dictate letters and just about any word...

                  Comment


                  • #10
                    Originally posted by Dominik
                    I'm not sure what the Honda stuff is used for, but i'd imagine that the reason it can understand almost anyone is because it has a ready defined command set, whereas the stuff we are using (well, i havent yet, but i'm planning to!) is used to dictate letters and just about any word...
                    What do you mean by predefined command set?
                    PowerVoice v1 | NaviVoice Source
                    GammaControl v2.4
                    SKINbedder v3

                    1995 Lexus SC300

                    Comment


                    • #11
                      As in, they probably only have to recognise Next, Previous, Vol Up, Vol Down etc. The MS Voice stuff is meant to be used to dictate letters etc (i think!).

                      Honda could have trained the voice rec. to listen to all sorts of voices and accents for each command. But MS has to learn how one user would say any possible word (hence the training program)

                      Comment


                      • #12
                        Originally posted by Dominik
                        As in, they probably only have to recognise Next, Previous, Vol Up, Vol Down etc. The MS Voice stuff is meant to be used to dictate letters etc (i think!).

                        Honda could have trained the voice rec. to listen to all sorts of voices and accents for each command. But MS has to learn how one user would say any possible word (hence the training program)
                        The Honda stuff is just ViaVoice... well, mostly...

                        And yes, there are a predefined set of commands that the computer will recognize. And yes, it will mess up sometimes, especially if you have a mouthfull of fries...

                        But the command set is fairly comprehensive, but it is to the point. Things like "XM Radio, preset 2" or "find nearest Chinese restaurant" (no joke). A basic and concise set of things to look for is the key... they can train the VR software to learn the most likely pronunciations of the keywords (find, radio, cd, track, etc) and then it would seem to be good "out of the box" (which it has to be, since I'm assuming patching that sucker is gonna be hard )

                        Finding an address is another thing... Most of the commercial stuff that IBM and AT&T have and they licensed is good though. If you call the MS activation hotline or AT&T directory services you'll hear that lady asking you to speak your answers -- and they catch almost 100% of the stuff and recognize it quite well IMO.

                        But until we have the money for those kinds of licenses, I'd assume that the on-screen keyboard will be our best friend...

                        Comment


                        • #13
                          Originally posted by viguera
                          Finding an address is another thing... Most of the commercial stuff that IBM and AT&T have and they licensed is good though. If you call the MS activation hotline or AT&T directory services you'll hear that lady asking you to speak your answers -- and they catch almost 100% of the stuff and recognize it quite well IMO.

                          But until we have the money for those kinds of licenses, I'd assume that the on-screen keyboard will be our best friend...
                          Recognizing a command set over a phone is much different then recognition in a car where your microphone generally is located at least 1ft from your mouth and subject to much more ambient noise. The free microsoft voice SDK will do just as well as anything from AT&T or IBM. I can easily get it to pretty near 100% if you let me control the configuration in which it will be rated.

                          Voice technology has been around in some form or another for over a decade and at this point, the core technology is pretty standardized between different vendors. You may have some features that make it easier to develop certain applications between each product, but there has been little revolutionary developments in voice technology that would make one product significantly better then another in terms of recognition ability.

                          Waiting for some vendors product with some exagerated claims or demonstrated performance in some limited test enviornment to be applied to carputers is just going to result in a lot of disappointment. IMO voice technology is now about as good as its going to get for a long long time and at this point the biggest shortcomings come from hardware used not from the software implementation.
                          StreetDeck.com Developer (I am Chuck)
                          Get StreetDeck at http://www.streetdeck.com
                          The Official StreetDeck Forums have moved, please visit us at http://www.streetdeck.com/forum for official support for Streetdeck.

                          Comment


                          • #14
                            Originally posted by Dominik
                            As in, they probably only have to recognise Next, Previous, Vol Up, Vol Down etc. The MS Voice stuff is meant to be used to dictate letters etc (i think!).

                            Honda could have trained the voice rec. to listen to all sorts of voices and accents for each command. But MS has to learn how one user would say any possible word (hence the training program)
                            I have a predefined command set.. NaviVoice.XML. I dictate letters only because I don't have a list of all the streetnames. I am trying to get something that the user will be able to put in the street names to have navivoice recognize them. And yes, it will be accurate. Same thing with numbers.. but I have to know the highest number to go up to. Also, do people say 8705 as "eight-seven-O-five" or "eight-seven-zero-five"?
                            Originally posted by viguera
                            The Honda stuff is just ViaVoice... well, mostly...

                            And yes, there are a predefined set of commands that the computer will recognize. And yes, it will mess up sometimes, especially if you have a mouthfull of fries...

                            But the command set is fairly comprehensive, but it is to the point. Things like "XM Radio, preset 2" or "find nearest Chinese restaurant" (no joke). A basic and concise set of things to look for is the key... they can train the VR software to learn the most likely pronunciations of the keywords (find, radio, cd, track, etc) and then it would seem to be good "out of the box" (which it has to be, since I'm assuming patching that sucker is gonna be hard )

                            Finding an address is another thing... Most of the commercial stuff that IBM and AT&T have and they licensed is good though. If you call the MS activation hotline or AT&T directory services you'll hear that lady asking you to speak your answers -- and they catch almost 100% of the stuff and recognize it quite well IMO.

                            But until we have the money for those kinds of licenses, I'd assume that the on-screen keyboard will be our best friend...

                            Originally posted by god_of_cpu
                            Recognizing a command set over a phone is much different then recognition in a car where your microphone generally is located at least 1ft from your mouth and subject to much more ambient noise. The free microsoft voice SDK will do just as well as anything from AT&T or IBM. I can easily get it to pretty near 100% if you let me control the configuration in which it will be rated.

                            Voice technology has been around in some form or another for over a decade and at this point, the core technology is pretty standardized between different vendors. You may have some features that make it easier to develop certain applications between each product, but there has been little revolutionary developments in voice technology that would make one product significantly better then another in terms of recognition ability.

                            Waiting for some vendors product with some exagerated claims or demonstrated performance in some limited test enviornment to be applied to carputers is just going to result in a lot of disappointment. IMO voice technology is now about as good as its going to get for a long long time and at this point the biggest shortcomings come from hardware used not from the software implementation.
                            Agreed SAPI is just as good IMO with predefined commands. ViaVoice and Dragon are good at recognizing context, but we don't need that.
                            BTW: Thanks for the xenarc bezel
                            PowerVoice v1 | NaviVoice Source
                            GammaControl v2.4
                            SKINbedder v3

                            1995 Lexus SC300

                            Comment


                            • #15
                              Ol33l, while you only use a predefined command set, SAPI still has to be able to understand every word (hence the training). Its just different slants on the voice recognition technology i guess.

                              Comment

                              Working...
                              X