I’ve been thinking about this for a while: years ago, I got my first laptop - an Asus Aspire, with 1GB RAM and an Atom processor. I got excited about the speech recognition built into Windows, and was impressed at its learning capabilities. 12 year old me could just take the tutorial and it would use that to learn my voice, and it got fairly accurate for a cheap mic with the sound of a hard drive and fan buzzing in the background.
What if Mycroft could learn your voice as you go, or had a skill that let you train it specifically for you? (“Mycroft, train my voice”). My little sister has to lower her voice before Mycroft clearly recognizes her. This same approach would work great for wake words - the ability to record a few examples of the wake word so it could recognize your own voice better, much like Alexa or Google Assistant.
There would be no need to upload it, since it is for the individual.
2 Likes
For wake words, there’s a skill for this already. For general STT, not yet. You can enable saving of voice clips. The training for improving STT is not possible at the moment, since the back end setup isn’t made for that. If you do your own Deepspeech or kaldi instance you could, but even that’s not fully automated.
And in the future as those two get easier to use, it might actually be fully automated. And that would be pretty cool.
3 Likes
That is helpful to know. Thank you!
I just started working on a new wake word using that. Don’t know how I never found the little guide.