Read the statement by Michael Teeuw here.
MMM-Hotword (v2) - whole-new-built
-
MMM-Hotword
MMM-Hotword is a hotword detector using snowboy.
You can use this module to wake another voice assistant or to give a command to other module.Screenshot

UPDATES
2.0.0 (2019-05-19)
- Whole new build-up
- Some annoying dependencies are removed.
- Installer is provided. (
installer/install.sh) - Personal model trainer is provided. (
trainer/trainer.sh) - Continuous recording after hotword detection is supported (Now you can say like “Computer, volume up” without pausing between
Computerandvolume up)- This feature could be used with
MMM-AssistantMk2 ver3.x(Not yet released, but will arrive soon) - Or you can use this feature with other AI or Speech-To-Text program.
- This feature could be used with
- Simple standalone commands could be available. (Without any Assistant or Speech-To-Text, you can make own voice commands with this module standalone.). Commands could be combined sequence (You can make “volume-louder” with voice models “volume” and “louder”)
- More universal models are added. (
computer,subex,hey extremeand more.) - Hotword detected could be displayed on screen of MM.
Download
[card:eouia/MMM-Hotword]
Last Tested; (2019-05-19)
- MagicMirror : v2.7.1
- Tested Environment :
- Raspbian Stretch (Raspbian 3B+) / node v8.16.0 / npm v6.4.1
- TinkerOS (TinkerBoard)
- Ubuntu 18.04 (NVIDIA Jetson Nano)
- OSX 10.14.4 (Apple MacBookPro) / node v11.12.0 / npm v6.7.0
-
Now you don’t need
MMM-AssistantMk2to make a simple command to handle MM. (But if you want, it’s your choice.) StandaloneMMM-Hotwordcan control MM with your voice simply. -
Hi @Sean installed without problems. Great rebuild of MMM-Hotword !
-
@Sean Tks for your great work job.
I use hotword to do a lot of thing with smart mirror. But still have a problem bellow:
I’d like to setup a lock function that hide all modun bellow a background screen and waitting only 1 hotword to show it again
Everything seem OK. But when locked, other hotword to interactive with modules still effect in background and lock function can’t prevent this
My question is How to temporary disable all hotword until it unlocks again? Tks! -
@smarthome
Well, hiding/revealing mechanism is only to callhide/showmethods of target modules. During its hiding, what to be done or not will be the responsibility of that target module itself. If some module should be not working during hiding, it is the business of that module. -
@Sean
Is there any “while…do” structure for using “hotword”, if have, we can do this lock function that temporary disable hotword
for example, the hotword to work with MMM Assistant MK2 is “google”
the hotword to lock is “lock_now”, the hotword to unlock is “hello123”
hotword lock detect procedure is background running
The structure is:
while { (("hotword lock detect procedure" )not equal to "lock_now" ) && ("hotword lock detect procedure" )not equal to "unlock_now" ) } do { - setup hotword "google" to interactive MMM Assistant MK2 } -
@smarthome
In some modules of mine (MMM-Clap and MMM-GroveGestures…) havecommandsMode or commandSetfor different command sets by condition, but not this. I think it could be considerable for your needs. But anyway, I’m quite busy nowadays, so cannot support it immediately. -
@Sean
Can I send double notificationExec to 2 Module with ony 1 hotword
1st to MMM Page-Selector
2nd to Assistant MK2 speaks about this action"open_news": { notificationExec: { notification: "PAGE_SELECT", payload:1 notification: "ASSISTANT_SAY", payload:"Open News page" restart:true, },Also combine
"open_news": { notificationExec: { notification: {"PAGE_SELECT","ASSISTANT_SAY"} payload:(1,"Open News page"} }, restart:true, },But in real test, only ASSISTANT_SAY can be executed
How can I do that for both?
Tks!
-
@smarthome
With onlynotificationExecof MMM-Hotword, you can send only one notification.
There are 2 solutions.- Use
MMM-NotificationTrigger, it can fire several notifications at a time by one trigger. - You can use
moduleExecto call inner methods directly.
moduleExec: { module: [], exec: (module) => { if (module.name == "MMM-Page-Selector") { module.notificationRecieved("PAGE_SELECT", 1) } if (module.name == "MMM-AssistantMk2") { module.notificationRecieved("ASSISTANT_SAY", "Open News Page") } } }(not tested code, but you can catch the concept. You can use
sendNotificationfrom sender module instead usingnotificationReceivedof target module)PS AMK2 has some TTS-like feature(ASSISTANT_SAY) but I recommend you to use dedicated TTS modules like MMM-TTS or MMM-GoogleTTS.
- Use
-
@Sean
1/Could you give me the code in MMM Notification Trigger2/Your code with ModuleExec didn’t work
3/I’m considering to use MMM TTS/Google TTS, but woud like to l try the simple solution first
-
{ module: "MMM-NotificationTrigger", config: { triggers:[ { trigger: "PAGE_SELECT_1", fires: [ { fire: "PAGE_SELECT", payload: (pl) => {return 1} }, { fire: "ASSISTANT_SAY", payload: (pl) => {return "ASSISTANT_SAY"} } ] } ] } },PAGE_SELECT_1from notificationExec of Hotword. -
@Sean
Your notification code didn’t work. Nothing happen :(
Could you check it again, plspayload: (payload) => { //OPTIONAL. transform received payload to what your target module wants.You mean that: MMM Notification transform only, can not modify the payload then transform?
-
@smarthome
Well, check the log. and tell me whether suspicious logs be there.fire: "PAGE_SELECT", payload: (pl) => {return 1}is equivalent to
.sendNotification("PAGE_SELECT", 1). whateverplwould come, it will be transfomed(or modified, whatever you want to call) and the fired notification would have1as payload. -
@Sean
Finally, it worked, many thank for your help! -
@Sean
I created some hotwords by myself on https://snowboy.kitt.ai/ under .pmdl extension, then save to model folder in MMM Hotword
After configure MMM-Hotword in MMM config file to use these hotwords, MMM-Hotword auto detect hotwords and do the action, why no input voice signal to mic, only noise…
I tried to incrase hotword from 1 word to 2 words, one to multiple syllables, but cant not stop this issuse
I’ve check the configuration, everthing is normal.
What’s happen? How to fix that?Tks!
-
@smarthome
Did you usetrainer.shof the module to make your pmdl? or made it on snowboy website? pmdl is somewhat sensitive so, if you made it on another platform (e.g. your desktop pc) or other device(e.g different mic with your RPI’s), it might not work.
Anyway, How do you figure out `only noise? -
@Sean
I have built two hot words like this to call each page. I call the first page works, but by the time I call the second page is not working? Please point out the error and help me!


-
@guongthongminh
Well, you don’t need to useMMM-NotificationTrigger. I can’t understand why you use that. MMM-Hotword itself has the feature of emitting notification(You’re already using.){ fire: "HOTWORD_RESUME", }, { fire: "HOTWORD_PAUSE", }It doesn’t make any sense at all. This means “HOTWORD_RESUME” and “HOTWORD_PAUSE” at a same time. Why this part be needed? This would make your MMM-Hotword being paused.
-
mmm-hotword is working well for me. I created it with first with the website Snowboy, later to make it better with trainer.sh. Sensity 0.5, later 0.9. With my voice it is working well, but for any noice from my wife, youtube movie, mmm-screencash, clocksound etc Google Assistant is activated many times. Also when nothing is said - only some “noise”. The file that is created (when it is Not the hot word) is only a small part of a word/ some sound - but it creates a connection with GA. Any idea how I can make it better?
-
Frankly says, not many options. There is always possibility of confusing with other sound sources(even noise)
Not surprisingly, real commercial products like google home also misbehavior sometimes. Nobody is saying g of google, but google home be activated.
However, not so may times. Because;- their detector is more dedicated to their own special pattern, which is result of advanced engineering and tech by google, amazon, apple,… ours are not such a level.
- they use arrayed mic with phonetic/acoustic science. So they can determine direction, distance and dimension of each sound sources. So they can KNOW what is real voice from human in the room, not noise, not music, not too far, and even not the sound from speaker itself at that moment.
Those are why they can detect precisely.
Hello! It looks like you're interested in this conversation, but you don't have an account yet.
Getting fed up of having to scroll through the same posts each visit? When you register for an account, you'll always come back to exactly where you were before, and choose to be notified of new replies (either via email, or push notification). You'll also be able to save bookmarks and upvote posts to show your appreciation to other community members.
With your input, this post could be even better 💗
Register Login