search

nyumaya / nyumaya_audio_recognition

Classify audio with neural nets on embedded systems like the Raspberry Pi
https://nyumaya.com
Apache License 2.0
82 stars 14 forks source link

training instructions? #25

Closed JarbasAl closed 3 years ago

JarbasAl commented 3 years ago

i would like to train a few models to evaluate this lib, are there instructions available?

I made a plugin for Mycroft, see https://github.com/JarbasAl/numaya_hotword , training would be really important for community adoption

yodakohl commented 3 years ago

Hi,

If you send me a list of the keywords I can train them for you.

best Regards,

Günther

Nov 27, 2020, 09:57 by notifications@github.com:

i would like to train a few models to evaluate this lib, are there instructions available?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, > view it on GitHub https://github.com/nyumaya/nyumaya_audio_recognition/issues/25> , or > unsubscribe https://github.com/notifications/unsubscribe-auth/AAGH6S2TO3FXLZPIMGZ5SODSR5SWVANCNFSM4UEUK5ZQ> .

JarbasAl commented 3 years ago

Hello

Thank you for the offer, it would be cool if you could train "hey mycroft", that way i could do some benchmarks against other wake word detector engines

However what i would like is instructions, users of the plugin linked above will want to train their own wake words, should i just point them to you? I'm more concerned with documentation than with getting wakewords for myself

Or is the training code closed source?

Thank you

yodakohl commented 3 years ago

Hi,

I can train hey mycroft, but the training code is closed source.

Best Regards, Günther

Nov 28, 2020, 13:15 by notifications@github.com:

Hello

Thank you for the offer, it would be cool if you could train "hey mycroft", that way i could do some benchmarks against other wake word detector engines

However what i would like is instructions, users of the plugin linked above will want to train their own wake words, should i just point them to you? I'm more concerned with documentation than with getting wakewords for myself

Or is the training code closed source?

Thank you

— You are receiving this because you commented. Reply to this email directly, > view it on GitHub https://github.com/nyumaya/nyumaya_audio_recognition/issues/25#issuecomment-735224056> , or > unsubscribe https://github.com/notifications/unsubscribe-auth/AAGH6SZQ3B4HYUQVIWF6VKDSSDSWLANCNFSM4UEUK5ZQ> .

JarbasAl commented 3 years ago

Thank you for the quick response, i will add that information to the readme in my plugin.

Will leave this issue open until you train "hey mycroft" wake word, but feel free to close it

Best Regards

JarbasAl commented 3 years ago

i humbly request yet another model for "wake up", that would be very useful to have

Thank you!

yodakohl commented 3 years ago

Added it to the list, eval model will be released together with the next version.

yodakohl commented 3 years ago

Eval models are available at http://nyumaya.com/models/V1.0.0_Jarabas.zip For easier performance evaluation there will be a fork of the wake-word-benchmark here. I'm currently pretty busy so this may take a couple of days before I can get to it.

JarbasAl commented 3 years ago

thank you for the models! So far i am quite impressed from the quick testing i made, i will do some proper benchmarking once i have some more free time. I haven't really looked into false activations, but when i say the wake words it seems to trigger almost always

keep up the good work! great performance and accuracy looks good