I actually did this 5 years ago. Called the one on my ACID box PeaterStein (repeater Einstein) I used AGI scripts and google STT/TTS and that was problematic then. Probably not much anymore. Once you have the converted speech, it's just a matter of conditionally evaluating text and executing it. Simple scripting really. The needed power is in the STT/TTS. And that is all external to the box. Requires internet access for the conversions. In the example video, it is sending speech to google, google sends back the text. You evaluate the text returned. Microsoft and a few others have somewhat open dev platforms for STT/TTS. It really requires less cpu than you think but will increase your internet bandwidth a bit depending on how you use it. The advantage with the newer tech is you can have a local device looking for your keyword without sending all the audio STT in monitor looking for it. I had to use a dtmf digit to pay attention to save the bandwidth. I'm sure you guys will have fun with it when he writes up a how2, but when I was experimenting, it became quite cantankerous as google developer criteria changed almost monthly and required either code re-writes or account changes or both. Ended up using MS STT and Google TTS. But it worked quite nicely. Had it tied to the wolfram db for specific info searches (another dev program). Lost interest in the whole thing. I find it better to just dtmf most of what I need as far as running the node. The things I was using it for on my asterisk box would not be allowed on ham radio. It's not going to do anything you can't do with just a script because you are just using the STT to execute your scripts. Example... I can press 3 dtmf's and get the same weather and a TTS engine speaks it for me. Compare that to speaking the keyword and waiting, giving it the actual voice command, and then wait for the response. I do still have a asterisk box on centos 6 using the old code. Haven't used it for years, may not work now because I doubt my google dev account works the same without a code re-write? We actually call these things interactive voice response systems. It's NOT 'AI'. To actually be AI, it would require that it could re-program it's own programming without human intervention. (change it's own code,not data, in a leaning process) The STT accuracy has to be much better now so you might not need to create much of a translate table for words that come back commonly spelled a bit different. But can be a issue if you don't keep it simple. The early days of speech recognition, we had to use a multitude of variations of the same audio and used the best match (fuzzy logic) in compare for results. Very sloppy but that was in the late 80's & 90's. When the first true STT started evolving in the later 90's early 00's, it required huge translation tables. Anyway, Have Fun, it's nice leaning experience for ya. I find no real value in it for 'repeater' use. Just a 'COOL' factor. Perhaps one of you will take it one step further ? Good Luck ! ...mike/kb8jnm On 1/6/2019 12:51 AM, Pierre Martel wrote:
Hi to every one.
After watching how VE4VR controled an IRLP node I started to wonder how could this be done on ASL. I dont want to steal the fame to this brilliant gentleman..
I am just damn curious how he did it and if it is done on ASL, preventing unwanted stuff to happen..
As they said there is more then a one way to skin a cat. here is my version..
I attached an image on how this could be done. But I also have some question...
AI setup.png As you can see I think that using the Chan_USRP driver to the Analog Bridge from the DVSwitch project there is a way to talk back an forth to the AI (PIcovoice project)
This would be a local node (1999) to the server and another node (28XXX) connected to the server by USBRadio or Simple USB or Chan_Voter would be the radio interface to the airwaves.
Now all this is nice and no one on the network links will say the wake work for the local repeater and all is well. NOT. It will happen that remote user will use the wake word and the AI will be all over the place. NOT GOOD. So beside using maybe another ctcss tone to enable the AI listening audio stream, wich could be a good idea as only people in the know would be able to trig the AI, is there another way differentiate the audio stream from a local one to a network one.. I imagine that we could use a small script to turn on or of muting on the audio stream to the AI?
Dont start trowing flames!! It is just for discussion and having fun waiting for the code to be released by William.. ( well I hope he will do it ;-) )
It is a damn fun time to be in ham radio!
Pierre VE2PF
_______________________________________________ App_rpt-users mailing list App_rpt-users@lists.allstarlink.org http://lists.allstarlink.org/cgi-bin/mailman/listinfo/app_rpt-users
To unsubscribe from this list please visit http://lists.allstarlink.org/cgi-bin/mailman/listinfo/app_rpt-users and scroll down to the bottom of the page. Enter your email address and press the "Unsubscribe or edit options button" You do not need a password to unsubscribe, you can do it via email confirmation. If you have trouble unsubscribing, please send a message to the list detailing the problem.
