Mycroft Community Forum

Custom wakewords R Us!

Sorry, sidetracked with some hardware issues. Added license and a new pull request.

1 Like

Will see if I can get you a model by Wednesday!

1 Like

Ok, thursday. Check https://github.com/MycroftAI/Precise-Community-Data/tree/amelia-model and see if that works ok for you?

Below doesn’t work , so i would like your help to create custom wake word .Hey K9 or hey canine.

                                                                     Regards,Geordy

{
“max_allowed_core_version”: 19.8,
“listener”: {
“wake_word”: “Hey K9”,
“threshold”: 1e-18
},
“hotwords”: {
“Hey K9”: {
“module”: “pocketsphinx”,
“phonemes”: “HH EY . K EY N AY N .”
}
}
}

Record a good set of sample words (20+, 50 better) from you and any other primary users, upload to the repo above and will see what we can do.

I may have done that completely wrong, Please check upload on github.

Not sure what you’re asking here.

Baconator, forked and uploaded files. 19 wav. I’m working on more today,

Baconator,
I’ve circled back to the wakeword, I pulled down the compressed tar file, but I’m having trouble extracting the tar file. I’m getting the following error:

[jnickel:~/Downloads] $ gunzip amelia-en-0.3.0-20191121-eltocino.tar.gz
[jnickel:~/Downloads] $ ls -l amelia*
-rw-rw-r-- 1 jnickel jnickel 89128 Dec 2 09:48 amelia-en-0.3.0-20191121-eltocino.tar
[jnickel:~/Downloads] $ tar xvf ./amelia-en-0.3.0-20191121-eltocino.tar
tar: This does not look like a tar archive
tar: Skipping to next header
tar: Exiting with failure status due to previous errors
[jnickel:~/Downloads] 2 $

Is there something special I should have done, or is the tar file possibly wrong?

Always try “file $filename” to see what it’s looking at it as.

Let me doublecheck…

Hm, I can download and extract it on two separate systems. Anyone else having issues?

Did another download and it appears I had a corrupted download earlier.

This model needs to be on the local filesystem of the device we’re working on. Should I put these 5 files in the .mycroft directory? And is the proper mycroft.conf settings like this:

“hotwords”: {
“amelia”: {
“module”: “precise”,
“threshold”: “1e-30”,
“local_model_file”: “/home/pi/.mycroft/precise/amelia.pb”
}
}

I believe that’s right, be sure to check the voice.log to see that it loads as expected.

As addition;

tar zxvf filename.tar.gz

Or, for modern versions of linux tar, just “tar xf $filename”

Hello Baconator
please check
https://github.com/Geordy6of9/K-9-voice-samples-en/blob/master/K9%20wav.tar

                                                                       Regards, Geordy

You fork that repo, then follow the instructions and upload to your own. After you’ve gotten that completed, submit a Pull Request and I’ll check it out.

So after a month of trying, I power up Mycroft this morning and he is not responding to Hey Mycroft. So rebooted him and try again 3 times. Then just thought what the hell, Hey K9. and he responded, I then asked K9 what is your listener, he responded with precise.

So K9 is working exactly the way i want…
Thank you, Did you do anything to make this happen?

                                                                    Regards, Geordy
1 Like

This looks great! but i need a bit more of explanation of how to use custom wakewords, do we need a set of files for each wakeworkd we want to use? or is it like a repository with many wakewords added?

As suggestion i think easy pronounceable names would be great to add, something like: Carl, Elvis, Mia, Ana or Danica or neutral names like Alex, River or Milan.

The problem with short names is they get false triggers far more often. Saying Hey xyz isn’t significantly more difficult than just xyz in any case.

Yes, you’d need to record a set (10+, 20 or more is better and 50 is great) of wav files with your wake word if you want to make a custom model.

1 Like