Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Aug 18, 2022 2:38:48 GMT -5
Finally I noticed that a speaker independent system like Elechouse's SimpleVR is in fact sensitive to background noise like the motors of my funfair which "destabilize" it while trains are in "pause" mode, but a speaker dependent system like Fortebit's EasyVR3 seems more reliable ; on the other side it needs training 7 words at least 2 times by each speaker (maximum 15 speakers). And I saw that other people had the same idea and developped (commercially) a similar system but dedicated to locomotive control. But since it runs on a tablet or a cellphone I think that it uses online recognition like Siri and Alexa. So, wait and see ! All the way a speaker dependent system would need a touch screen to allow training "on site", otherwise the control panel would look like in a spacecraft, and there would not be enough input-outputs on my processor. (BTW Rokuhan sells a power supply remotely controlled via an app, but without speech recognition functions.)
|
|
|
Post by markm on Aug 19, 2022 12:06:23 GMT -5
Note that the Rokuhan Apple app is at risk of being removed due to developer inactivity. So, if you think you might want to use it, download it now.
You journey through voice recognition has been interesting and relevant for those of us who start the day with: "Hey Siri, Good Morning" to bring our house alive. And it still makes mistakes.
As far as training, I think anything beyond "Good morning, my name is Mark." is too much. Any training needs to look like normal interaction.
If you can find a system that can limited accepted responses, like numbers or single action words, you would probably have more success with speaker independence. This avoids training and is how the corporate call-center programs generally work.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Aug 19, 2022 14:41:36 GMT -5
This avoids training and is how the corporate call-center programs generally work. All these systems (call-center or home control) run on mainframes, not on microcontroller-size systems. For example, the well known Alexa sends Your voice (I think, coded in VoIP) to a central computer (located at Amazon, I presume), but hackers can connect to Your microphone combination too and so You will be "watergated" ... In SI mode, SimpleVR is the "less bad", but very sensitive to noise (especially coughing), they say one should use commands as long as possible (and finally direct manual commands will be faster ! ). As I wrote, EasyVR3 in SD mode is less sensitive to background noise ; the French modeller who controlled his locos via Arduino (and DCC , several locos can run simultaneously and their motors are not silent) with some functions like the Lionel system used this. And I think that Lionel uses the SI remote decoding (SIRI for Apple and the equivalent on Google for Android), but it has some latency time. With EasyVR3 the recognition process is very fast (there is a "Fast SD" option) but every user (15 possible) has to train first the seven keywords (4 for trains, the railcar counts for two because it can run into 2 directions, and 3 possible movements), each of them 2 (at least) times ; multiply ... And since the system is intended to be used mainly by kids (or a few nerds), I think I will sleep several nights more before I decide to implement such a "gaseous plant" !
BTW I am absolutely not interested in using a cellphone ... especially an Apple model ! The price is somewhere between 3 and 4 AZL locos ...
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Aug 27, 2022 1:27:11 GMT -5
About voice recognition : the EasyVR3 (like most other similar systems, I presume), connected with the main control board of the layout, needs in fact that the "speaker" makes a break before each keyword, if there is no feedback (led or beep, which becomes quickly boring). Now go and explain this to kids ... I wonder if it is a really usable solution, excepted for demonstrations made by ... myself ! And in this case I can train the system connected with my desktop computer, and don't need on the layout a training subroutine or user selection (excepted perhaps for the language select, between English and French ; but since it is a US layout, I could be satisfied with English ! ).
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Aug 29, 2022 13:03:10 GMT -5
And finally the module EasyVR3 had a system issue, memory corrupted, locked for at least 15 seconds, etc ... Long live pushbuttons !
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Aug 30, 2022 5:04:48 GMT -5
How to lose Your voice after 15 minutes maximum :
For the turntable and the turnouts they use apparently names like Alpha, Bravo, Charlie (I heard Romeo for turnouts) ... For the locos Franklin, Odin, Thomas, etc (!). I took a glance at their homepage, they claim that their system is speaker independent ... On the other side the vocabulary is rather "strange" and not really related with model (or prototype) railroading ; finally it is rather like some codes used by air traffic controlers, or even in movies ... by spies ! It's almost as if I would control my layout by voice ... IN BINARY ! (it's a provocation ; it would in fact be very easy, 2 bits for the train and two for the itinerary, but it would be really stupid ! ). BTW I am surprised how many such systems are available commercially (look on YouTube for "model railroad speech control"). After all, if they succeed in selling them ...
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Sept 4, 2022 2:09:41 GMT -5
I spoke with a modeller friend about this system (BTW it uses my "beloved" DCC), and he said that such controls would lack of interest as long as one must use names for locos like Odin, Franklin ... instead of "Br01" (a German Pacific). And I tested a workbench simulation (with Leds and without background noise) of my "binary" control, it could work, but saying a combination of 5 successive "alpha" and "omega" is rather suitable for a Sci-Fi movie ! So, since there are no OFFLINE systems for natural language (like online recognition, Siri, Alexa, etc, which run on mainframes) or at least railroad related keywords, and this seems to be in a far future, manual controls (via switches or perhaps a touch control screen) will be the best solution ...
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Sept 11, 2022 21:51:01 GMT -5
The final comment about this subject, found on a German railroading forum (approximate traduction) : " Hello, it's bad enough that people just still look at their phone and talk to him too. We model railroaders shouldn't start with this nonsense. If someone has nothing to talk to, he should get a dog. " Finally I agree with this. And a friend of mine said that kids like touch screens and use only switches if there is no other system available. It's time that they learn again the contact with real, interlocking switches, with springs and where the "click" is not only generated by a loudspeaker ...
|
|
|
Post by zman on Sept 12, 2022 6:09:22 GMT -5
I agree with you 100%! About 18 months ago our club built 8 T-Trak modules (N-scale) designed with kids in mind. Ergo, visually appealing to youth. The layout is DC operated and duel track. We have trains running on both tracks. We have the power stations located at the end of the table so the kids (sometimes adults) can get a hands on opportunity to run trains. It has been a successful draw at train shows and displays that we have used these modules.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Sept 13, 2022 1:40:55 GMT -5
Without beginning again a discussion about voice recognition, which I "burried definitively" (all the way there is at least presently no reliable system but the recognition of Windows : I tested 6 modules from 3 sources ! ), and admitting that pushbuttons are a really fine thing, consistent with the era, I must confess on the other side that a touch screen (Nextion capacitive) would have a great advantage : modifying the control panel and adding feedback functions without modifications to hardware. For example, only modifying the aluminium panel (look at previous page) costs more than 80 Euros. And You will notice that there are no feedback lights for the selected train and itinerary ; these are useless in manual mode, since everyone can see the buttons which are set (due to the interlocking function). But You will perhaps have noticed too the black button at the right (labelled "auto" ), which puts the layout in automatic mode (randomly selected trains and itineraries) : in this mode it's not possible to see directly neither which train will be started next, nor the itinerary it will perform, nor the time remaining before start. But this could be interesting for visitors waiting in front of the control panel while I walk around showing details of the layout (the random mode was designed for this purpose). Adding control lights would involve "deep" modifications to the panel and adding connections to the processor card ... so, making a new PCB ; but with a touch display it would only need "free" software modifications (a Nextion is only connected via 2 pins in serial mode). I am scared of the number of modifications to come which will be suggested by my visitors : "It would be fine if You could add this function, etc ... " (for example, look at Mark's suggestion on previous page, which would need a display, despite of the fact that in conjunction with the operating mode it is rather useless in my eyes). So I will finally wait if my visitors make really interesting and "reasonable" suggestions, and only then I would migrate to a third and last panel (cost about 200 Euro, since I already bought the Nextion for testing), it would be cheaper than all the other successive modifications probably suggested in the future ! As I wrote, I prefer the present version and will keep it as long as possible ...
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 10, 2022 9:46:31 GMT -5
Finally I ordered the (third ! ) front plate for the touch display (Nextion 5" capacitive) at my carpenter, this will be the final hardware modification ; all other additional functions will be performed via modifications at software level. I have already programmed the user interface, it runs in "simulation" mode with an evaluation board and works fine ...
This is a screen hardcopy from the editor, it's not possible to take a picture due to the backlight ; so the names and the borders of the "buttons" are visible.
|
|
|
Post by BAZman on Oct 20, 2022 11:35:01 GMT -5
We in the America used cheap USB Android tablets and PanelPro (Part of DecoderPro www.jmri.org/help/en/html/apps/DecoderPro/index.shtmlYou load a graphic (layout) and place objects (icons) buttons, toggles, etc from their library or make your own. Then, attach the function/action for that icon (turnout/switch, start/stop/, slow/fast, etc). You still need a computer (any platform) It’s Free, open source.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 20, 2022 13:50:45 GMT -5
The hardware costed 95 Euros, including taxes, shipment and first of all a mounting frame with a rubber seal. The configuration software can be downloaded for free and creating the interface looks much like using a RAD development language similar to VB(A) or Delphi, for example : clicking on a pad triggers the transmission of a string, including the ID of the activated element. And the same way You can select the active screen (via a command string), change the icons displayed on the buttons and even use a "bargraph" to display the remaining time before start when the system is in automatic mode. There is a loudspeaker which can play wav files under software control (but I won't use them ... ). And finally the only interface needed on the board is a TTL level serial one at 9600 Bauds, so I think that for me an ATMega8515 (in worst case a 162) will be enough and manage all operations, including driving the trains. On the other side I wonder if Your system is not only dedicated to DCC controlled layouts ?
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Nov 12, 2022 1:51:31 GMT -5
This post about using Kittenbot ASR module has been deleted ; I THOUGHT honestly that I had found the perfect speech control system, but finally Murphy stroke again : while testing various "speeds" of speaking I found a word, not included in the vocabulary : "red" , followed by "light on" after a too long time lapse acts like "red" alone and can so be interpreted as "rest", and this puts the module into the idle state without any feedback, not even the sound related with timeout. And You are surprised to get no more reaction from the system, until You understand and say again "Hello Kitty" for wake up ... One more disappointment for the "Nerd" ! All the way this function of "wake up" is absolutely stupid in my application, data sent by a speech command module are ignored and rejected by the sequencer while a train is running. But there is no parameter to disable it on the Kittenbot.
|
|
|
Post by markm on Nov 12, 2022 23:30:06 GMT -5
Don’t feel too discouraged, I still have to argue with Siri to turn my lights on and off.
|
|