Problem with training Raven

I am trying to create multiple recordings so you are more likely to recognize my voice when I call my assistant (I guess). What I do is create 3 recordings with the system (on the web) and then change their direct name in the profile (via SSH), and so on more than 20 times.

When I did it and called her, she didn’t understand me, nothing happened, when I deleted all the recordings and only left 3 by default, she did. I am doing something wrong?

Contrary to Mycroft Precise than needs loads of keyword samples (the more the better), Raven only need 1 to 3 recordings. Adding more recordings will only negatively impact accuracy and CPU usage.

Try to make your recordings in the most quiet environment possible. You can review your templates with Audacity to remove clicks and pops if any.

Hope this helps.

That’s right, it helped me, I’ll read more about Raven. I would appreciate any other additional information you can give me.

I have another question, if I record my voice in Raven (3 sounds), will it be recognized from a distance? Will you recognize when someone else says that word or only when I say it?

It should recognize your keyword from a distance (you can adjust the detection thresholds as needed).

As Raven is based on templates of your voice, it can only recognize your voice.

Do not mix multiple voices in the same keyword it will backfire badly (lots of false positives).

If you need to recognize a keyword for multiple voices, simply use different keywords (one per voice, can be the same pronounced word) with 1-3 templates each (beware of the CPU cost).

Hope this helps.