|
|
Post by lemiceterrieux on Jul 17, 2024 14:33:18 GMT -5
Some commercial tests have been made into this direction, often PC and DCC based : look here in my former thread ("How to loose Your voice after 15 minutes"). Don't disregard the Lionel system, which seems to work with Bluetooth and Android (and iOS ? ) And even Alexa-based systems exist on YouTube. In fact my layout is macro controlled : one must only select the train (3 but in fact 4, because the railcar can start in both directions, and there are 2 freight trains) and the type of itinerary (presently 3, maybe 4 in a near future), press the "Start" key and the sequencer controls the complete action ; only one train can run at a time. So I try to replace the keys by voice commands, decoded by a dedicated (offline) module (I have been working in IT for 17 years in the past). I can say that I tested most of them, even some rather expensive (and/or slow) versions ; finally my choice kept only two of them : the EasyVR3+ from Fortebit-Sensory, and the Chinese VC-02 module made by Ai-Thinker in Shenzhen. Presently I make desktop tests with a STK200 board (Atmel based, programmed in Pascal).
The first model works well in speaker dependent mode (like ... Windows 7 ! ), but every user has to "train" it , i.e. repeat twice the 8 or 9 command expressions ; it works less well in speaker independent mode, especially in other languages than English, despite of the fact that the software to create the vocabulary is rather expensve (3/4 the price of a loco). Sometimes it seems to "hesitate", one could think that the command was not understood and, just at the moment when You try to repeat, the answer comes (the module has a built-in sound player). The second model is much less expensive, less than 10 Euros or Dollars, available from several sources (even in USA). It works in speaker-independent mode, answers very quickly but can only "understand" English (and Chinese). If You choose carefully the commands (like "freight train", "railcar left" or "continuous") it is very reliable. But it needs an external feedback, either control lights or speech via the Adafruit Wav-player. The module has a bootloader, and the software to create the vocabulary works online and this is the weak point : the server is sometimes crashed and You have to wait tilll it runs again. But finally I think I will use the second module ; after creating the vocabulary (and downloading the corresponding configuration file to my PC) I don't no longer need their server ; and the future visitors who can't speak English can still use the control keys ...  |
|
|
|
Post by lemiceterrieux on Jul 30, 2024 0:01:31 GMT -5
About VC-02 : it would work fine ... but not concerning the "audio" wake-up function ! "They" write that it is possible via hardware, but I didn't find the corresponding function on the configuration page. And the audio sensitivity for wake-up is worse than for commands : depending on the position of the microphone You need sometimes to shout the corresponding expression to get the module awaken. I mailed to them but without too much hope : I read on another forum that a user had the same issue and never got an answer.
|
|
|
|
Post by lemiceterrieux on Aug 1, 2024 5:14:31 GMT -5
Today I got an answer ... but not exactly to my question ! I explained that it was not satisfying and why this didn't solve my problem ...
|
|
|
|
Post by lemiceterrieux on Aug 3, 2024 9:28:17 GMT -5
I am still waiting for a SATISFYING answer from Ai-thinker or their US-dealer, Digikey. I think it's stupid that at the beginning the system says : Please say "Hello system" to activate. And I must answer "Hello system" perhaps several times, before I get the answer : Ready Or should I use "Hello Casey Jones" ?  Both are equally stupid ... like Alexa. If I don't get a solution till the end of next week I will consider the speech control as dead, buried and forgotten ... and this thread as closed !
|
|
|
|
Post by lemiceterrieux on Aug 4, 2024 4:14:02 GMT -5
I had issues with the command "special", whose purpose is to make the freight train and hoppers convoy switch in and out between the dead end tracks and the "secundary" track in the station (without running on the main line), and makes the railcars run automatically back and forth between the town and the village, with a pause. I replaced it with "specific" (it fact it has a "specific" function, not the same for all trains), and now I can use the "wake-up-free" mode with ALL my commands, the first command spoken activates the module, too ; "special" worked as a command once activated, but not for the activation itself at the beginning. And so I need no longer any dedicated expression for this purpose ...
|
|
|
|
Post by lemiceterrieux on Aug 9, 2024 23:15:24 GMT -5
Since I rejected the Nextion touch display, I will use again the Adafruit sample player for audio feedback ; and to avoid stupid questions why there is a female or male voice (created with ttsmp3.com) I used a "robotic" voice made with SayIt (parameters 1013 9 0 4). Sounds horrible, doesn't it ? But in fact exactly as in some B/W SciFi movies around 1960 (the approximative era of my layout ).
|
|
|
|
Post by tikigrotto on Aug 10, 2024 2:59:43 GMT -5
I was serious about laying off the caffeine
|
|
|
|
Post by lemiceterrieux on Aug 10, 2024 3:30:29 GMT -5
Since Lionel offers such a system commercially (without spoken feedback, because all commands are executed immediately), I think it should be interesting for some "freaks" ... And, concerning the voice, what do You think ? It would be as easy to use a voice from ttsmp3.com ... But, which ? I think there would always be people to criticise the choice, either male or female. In worst case, I could add a manual selector, so everybody could be satisfied ... My final choice : the trains by Kendra, the itineraries by Matthew, and the "system messages" by the robot !
|
|
|
|
Post by lemiceterrieux on Aug 10, 2024 15:54:19 GMT -5
And now bad news : finally the "wake up" method which I wanted to use (by the commands themselves) was not completely reliable, so I went back to "Hello system" ... which was again not reliable, too. I think that the problem is perhaps related with reset issues of the VC-02. So, instead of wasting my time with testing a system which will only lead to laughters from my visitors, I stop developping the speech control with the VC-02. Maybe I will test again the EasyVR3+ with training ? I read about a french modeller who uses it (together with Arduino and DCC  ) ; at least, that proves that I am not alone in making such tests. But this would only be suitable for my own usage in a demonstration context, not for most visitors. Or I stop until I find a REALLY RELIABLE system. But I wonder if this will ever happen ? WOMBAT = Waste Of Money, Brains And Time !
|
|
|
|
Post by lemiceterrieux on Aug 11, 2024 9:54:38 GMT -5
I tested EasyVR3 in speaker dependent mode, one would need to train again all 9 commands (twice) very often, because I noticed that voice can change very often too. And some initializations can fail. On the other side, it is very fast and one can use FRENCH. But in speaker independent mode, it is too slow, sometimes one thinks that a command has not been recognized and finally the confirmation comes ...  |
|
|
|
Post by lemiceterrieux on Aug 12, 2024 3:29:10 GMT -5
Finally I think that the only reliable system is the "Hello Kitty". Ok, it understands (red/green/blue/yellow) light on, this would allow the control via a tree-structured menue, but saying "Hello Kitty" (or "Hello kittenbot", so I could use as logo the "Mechano" robot from a "Tom and Jerry", 1952 ... ) before each group of commands would be quickly felt as boring (despite of the fact that people consider as normal the expression "Alexa" in the same context). But this would need a control panel with feedback lights (the digitasts ? ) Wait and see ...
|
|
|
|
Post by lemiceterrieux on Aug 13, 2024 0:13:10 GMT -5
The "Hello Kitty" system (in fact, You can find the module here) would operate well, but would lead model railroad control to a rather ... silly context in relation with the periodic wake-up commands (Hello Kitty / Hello kittenbot). On the other side, it could be funny. Don't forget that even the Japan National Railways decorated a Shinkansen with "Hello Kitty" themes  (this version has been sold by Rokuhan for a while) and another one with "Neon Genesis Evangelion" (  ; that's only my own opinion). Concerning the EasyVR3, I noticed meanwhile that it can be triggered by background noise (especially by my breathing, when I use a headset). I added (in speaker independent mode) a function which said when an expression had NOT been recognized and got a volley of false triggerings. Not useable ! Concerning the "Hello Kitty", I say : Wait and see ... |
|
|
|
Post by lemiceterrieux on Aug 18, 2024 9:12:43 GMT -5
And I noticed finally that neither "Hello Kitty" is very reliable, sometimes I had to SHOUT several times before the system answered ; it is very sensitive to the position of the microphone (I connected an external version). But meanwhile I THINK that I found a solution with the EasyVR3 in speaker INDEPENDENT mode (!) and in FRENCH, perseverare diabolicum est (until I find another issue ) ; "I see the light at the end of the tunnel" (from "Starlight Express" ) ; a hands-free microphone and spoken feedback via loudspeaker (or a combined microphone-earphone headset), no feedback for non-recognized commands and 3 reduced vocabularies according to context : Phase 1 : "Please select the train" . You have the choice between 4 trains ("Railcar" is 2 trains, right and left) . Verification if train can run. Phase 2 : Selected train is spoken, "Please select itinerary" . You have the choice between 4 itineraries, or back. Itinerary is verified. Phase 3 : Selected train and itinerary are spoken, "Train can start" . Back or start. And the wake-up is done one time only, and by SOFTWARE. OK, it's a "gaseous plant" , but it works well ... at least on desktop ! Wait and see ...
But before I go on, I wanna verify that the noise of the motors of my funfair doesn't interfere, and test the system on desktop with (grand)children of my neighbours, i.e. the voices of YOUNG girls and boys (my age is 71, so the voice is probably a little bit different ! ).
|
|
|
|
Post by lemiceterrieux on Aug 21, 2024 15:25:10 GMT -5
PS : some tests brought to me these conclusions : - NO BACK OPTION, otherwise You get very quickly lost in the tree structure. When the train and the itinerary have been defined and if there is no issue, train will start immediately without any additional command. - NO SPOKEN FEEDBACK for the selected train and itinerary ; be confident in the reliabilty of the system ; all the way the software avoids accidents, so it is not severe if a command is misunderstood. - ONLY beeps or single word feedbacks, otherwise You will feel as with some interactive phone systems (If You wanna ..., press key 1, etc ) |
|
|
|
Post by lemiceterrieux on Aug 26, 2024 23:24:36 GMT -5
The EasyVR3 produced an error today, confusing 2 commands : in French, I said "continu" and it understood "double" (?). On the layout errors can't cause any accident, the software preventing from all "dangerous" inputs via the keyboard, too. In worst case it would have triggered some laughters. All these systems are sold with an accuracy of 95 tp 98 percent. I wonder how safe such controls like Alexa are ? In the "IronMan" movies home control is perfect, but is it in real life, too ?
|
|
Both are equally stupid ... like Alexa. If I don't get a solution till the end of next week I will consider the speech control as dead, buried and forgotten ... and this thread as closed !
) ; at least, that proves that I am not alone in making such tests. But this would only be suitable for my own usage in a demonstration context, not for most visitors. Or I stop until I find a REALLY RELIABLE system. But I wonder if this will ever happen ?
(this version has been sold by Rokuhan for a while) and 
; that's only my own opinion). Concerning the EasyVR3, I noticed meanwhile that it can be triggered by background noise (especially by my breathing, when I use a headset). I added (in speaker independent mode) a function which said when an expression had NOT been recognized and got a volley of false triggerings. Not useable ! Concerning the "Hello Kitty", I say : Wait and see ...