Custom wakewords R Us!

Archive - Mycroft Project Archive - Machine Learning
41

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

42

As addition;

tar zxvf filename.tar.gz

43

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

44

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

                                                                       Regards, Geordy
45

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.

46

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
47

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.

48

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
49

@baconator

Are you still doing this? One of the features I wanted from the beginning was ‘Hey Jarvis’, as a Precise model. I have been really tempted to record the wake word training. I probably could get around 6-7 people as samples. Parents in their 70’s, men and women in 30’s-40’s, and at least one super awesome 4 year old nephew. The only catch would be that a lot of the recordings would most likely have a country twang to them. I personally try to speak with no accent, but I am surrounded by country folk. Is this doable?

1 Like
50

accents, different ages, female and kid’s voices are all great.