As I wrote in other threads, I planned to use a voice recognition to control my railroad layout, a keyword should have been "start now" ... until I noticed that the same effect would be triggered by "fart now" ; this is no suitable way to impress kids who use smartphones with voice recognition ! Maybe that nothing is better with some people than the good old pushbuttons (digitasts) ... and that life of developers can sometimes be very funny ! :)
There is always the possibility of using a HIDDEN switch to activate or not the voice system ... and using "departure" instead of "start now". And I think I will often use it in "OFF" position, depending on the visitors !
:)
Last edit: Bertrand BAROTH 2021-05-03
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Try a free speech to text converter https://www.folio3.ai/blog/best-free-speech-to-text-software/
Then check the text string against the text you want. I did have BBC basic code one time that checked input text against a correct answer and allowed for a set number of differences to allow for spelling mistakes in the input text.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
In fact I tested "Vocals", which works fine to detect keywords, using the built-in speaker dependent voice recognition in Windows and creating keyboard shortcuts which could be interpreted by a Windows prog, perhaps written in Delphi (I succeeded in controlling a 1/24 RC model tank this way, interfacing the transmitter via USB). Unfortunately my desktop computer is not in the same room as my model railroad ; and about Android and IOS, I have NO SMARTPHONE, only a "basic" portable telephone, for calls and SMSes. So I was looking for a module which could be "embedded" into my command console and connected to my ATMEGA board via serial, spent "much" money ... for finally nothing !
:-(
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Searching for "free voice to text" a quad core android... from wish.com...joking, would be ok or a raspberry pi zero.
Seems some software needs an online connection but only the hardware to sample the sound and send it to the internet and the results are sent back.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
And I don't have an Internet connection in my railway room, and no WiFi ...
About speech recognition on the Web :
1) I read some negative posts about Google recog.
2) I called a "service" on the net, where I had to dictate an identification code character per character, I think they use a "big" computer or an online service. After 3 times I got a human operator on the line, the first thing he asked me was ... to say my code. No comment !
Last edit: Bertrand BAROTH 2021-08-08
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I remember my sinclair zx spectrum and a simple program to "sample" the mic socket used for loading tapes. You could do 1 bit sampling .. ie feed audio in and was it over a voltage and it was hi or low, which was stored in ram and could be played back and the sound was recognisable.
Maybe do this with a pic and compare the 1 bit sample with a new one allowing for so many differences.
Maybe ok for 2 words depending on memory. Just an idea.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
A similar method was popular in the 80's - 90's, but one needed not only a rough sample, which would have been too dependent of speed ; in 1993-94 I built in fact a model boat controlled via RC this way (with a 6500 family processor), it was even published in a model boat magazine (in France, RC-Marine #42) ; but I used numbers of occurencies of mean frequencies, inspired by the book "How to build a computer-controlled robot", by Tod Loofbourrow from 1978, using the Kim-1, still available at Amazon (for more than 70 $ ! ). Assembly was performed on C64 with SM-MAE64. It worked ... but only for me, nobody else was able to "repeat" words he trained before with sufficient accuracy ...
:(
I guess Alexa (or whatever name is) is a simple device that relies on the internet servers for doing the computer stuff.
If I was you, I would have abandonné the voice command.
c'est un mal de tête :)
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
In fact, I gave up voice recog for my railroad ... I can say I tested EVERYTHING, possible ... and impossible. What bothers me is that some services still rely on it ... and even reject You if they don't "understand" what You say !
The only thing I will perhaps do is replacing the "modern" digitasts with more "vintage" switches (but keeping the "invisible" processor control) : the era of my layout is 1965, even before Amtrak creation.
Last edit: Bertrand BAROTH 2021-08-08
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have more unfinished gcb projects than I can wave a stick at :)
translates as --- J'ai plus de projets gcb inachevés que je ne peux agiter un bâton à
bonne chance avec d'autres idées
GCB is good for a free program but can not do voice recognition. well.. not out of the box...
needs some imagination and research. Maybe a dead end... maybe not. Try another project?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
And it works anyway ! With Elechouse's V3 (speaker dependent), two vocabularies each with 4 carefully chosen words, activated according to context ... But "start" is still triggered by ... (Bart Simpson). With Easy VR3, one needs to insert delays (40 to 50 ms) into serial transmission, but these are NOT documented in the manual ...
:)
Wait and see if the same trick works with Elechouse's Simple VR (speaker independent, it would be too good ! ).
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
As I wrote in other threads, I planned to use a voice recognition to control my railroad layout, a keyword should have been "start now" ... until I noticed that the same effect would be triggered by "fart now" ; this is no suitable way to impress kids who use smartphones with voice recognition ! Maybe that nothing is better with some people than the good old pushbuttons (digitasts) ... and that life of developers can sometimes be very funny ! :)
There is always the possibility of using a HIDDEN switch to activate or not the voice system ... and using "departure" instead of "start now". And I think I will often use it in "OFF" position, depending on the visitors !
:)
Last edit: Bertrand BAROTH 2021-05-03
The final conclusion, after several months of testing several modules, I would say : LONG LIVE OLD FASHIONED MANUAL SWITCHES !
:(
:-)
Try a free speech to text converter https://www.folio3.ai/blog/best-free-speech-to-text-software/
Then check the text string against the text you want. I did have BBC basic code one time that checked input text against a correct answer and allowed for a set number of differences to allow for spelling mistakes in the input text.
In fact I tested "Vocals", which works fine to detect keywords, using the built-in speaker dependent voice recognition in Windows and creating keyboard shortcuts which could be interpreted by a Windows prog, perhaps written in Delphi (I succeeded in controlling a 1/24 RC model tank this way, interfacing the transmitter via USB). Unfortunately my desktop computer is not in the same room as my model railroad ; and about Android and IOS, I have NO SMARTPHONE, only a "basic" portable telephone, for calls and SMSes. So I was looking for a module which could be "embedded" into my command console and connected to my ATMEGA board via serial, spent "much" money ... for finally nothing !
:-(
Searching for "free voice to text" a quad core android... from wish.com...joking, would be ok or a raspberry pi zero.
Seems some software needs an online connection but only the hardware to sample the sound and send it to the internet and the results are sent back.
And I don't have an Internet connection in my railway room, and no WiFi ...
About speech recognition on the Web :
1) I read some negative posts about Google recog.
2) I called a "service" on the net, where I had to dictate an identification code character per character, I think they use a "big" computer or an online service. After 3 times I got a human operator on the line, the first thing he asked me was ... to say my code. No comment !
Last edit: Bertrand BAROTH 2021-08-08
I remember my sinclair zx spectrum and a simple program to "sample" the mic socket used for loading tapes. You could do 1 bit sampling .. ie feed audio in and was it over a voltage and it was hi or low, which was stored in ram and could be played back and the sound was recognisable.
Maybe do this with a pic and compare the 1 bit sample with a new one allowing for so many differences.
Maybe ok for 2 words depending on memory. Just an idea.
A similar method was popular in the 80's - 90's, but one needed not only a rough sample, which would have been too dependent of speed ; in 1993-94 I built in fact a model boat controlled via RC this way (with a 6500 family processor), it was even published in a model boat magazine (in France, RC-Marine #42) ; but I used numbers of occurencies of mean frequencies, inspired by the book "How to build a computer-controlled robot", by Tod Loofbourrow from 1978, using the Kim-1, still available at Amazon (for more than 70 $ ! ). Assembly was performed on C64 with SM-MAE64. It worked ... but only for me, nobody else was able to "repeat" words he trained before with sufficient accuracy ...
Last edit: Bertrand BAROTH 2021-08-08
:(
I guess Alexa (or whatever name is) is a simple device that relies on the internet servers for doing the computer stuff.
If I was you, I would have abandonné the voice command.
c'est un mal de tête :)
In fact, I gave up voice recog for my railroad ... I can say I tested EVERYTHING, possible ... and impossible. What bothers me is that some services still rely on it ... and even reject You if they don't "understand" what You say !
The only thing I will perhaps do is replacing the "modern" digitasts with more "vintage" switches (but keeping the "invisible" processor control) : the era of my layout is 1965, even before Amtrak creation.
Last edit: Bertrand BAROTH 2021-08-08
I have more unfinished gcb projects than I can wave a stick at :)
translates as --- J'ai plus de projets gcb inachevés que je ne peux agiter un bâton à
bonne chance avec d'autres idées
GCB is good for a free program but can not do voice recognition. well.. not out of the box...
needs some imagination and research. Maybe a dead end... maybe not. Try another project?
And it works anyway ! With Elechouse's V3 (speaker dependent), two vocabularies each with 4 carefully chosen words, activated according to context ... But "start" is still triggered by ... (Bart Simpson). With Easy VR3, one needs to insert delays (40 to 50 ms) into serial transmission, but these are NOT documented in the manual ...
:)
Wait and see if the same trick works with Elechouse's Simple VR (speaker independent, it would be too good ! ).