Improve wake word model

[MXC48]

The biggest problem at the moment with this feature is that I'm French 😅 so I have a French accent when I pronounce "hey Dicio" and the app doesn't recognize the phrase and therefore doesn't wake up...
We'd have to set up a database to collect data on the pronunciation of the sentence in different languages to be able to train the model.

[sandoxy]

When I speak (not native speaker) wakeup-word works 1 to 20, so not well. But it's a great feature, thanks!

Maybe, I pronounce dicio the wrong way:)

[Handrail9]

Native English speaker here, it also works about 1 in 20 times for me as well lol

[paolo-caroni]

I don't want to advise closed service....
But use a famous translator online, write "ei dicio" in latin language and use the listen function. I know that latin is a dead language, but dicio is latin, and most language are latin derivation/intersecation. French and english included :-)

[sandoxy]

Thank you for the advice. I did use a famous translation service. When this service pronounces the wake up word, it works almost every time. When i try to repeat it the same way, it doesn't. By the way, i am a musician and my abilities to hear and repeat are not the worst:) It just seems to be not very tolerant, the pronunciation must be perfect.

[paolo-caroni]

So the idea of @MXC48 of improve the hey_dicio_v6.0.tflite model would be important to different audiences. Maybe @Stypox can collect audio from users here on repo to improve the model.

[HyperCriSiS]

I think best would be if an own custom wakeword could be trained. 😀
I am not a fan of fixed wale words. 😕

[paolo-caroni]

@HyperCriSiS see #238