training model using non-english positive and negative samples.

[Naif-Ajeeb]

i tried using the simplified version of the automatic model training notebook, i got the model and i noticed lots of False Positives (detects the word when i didn't actually say it). i thought the problem might be from the negative dataset, because the wake word is in arabic and so when i speak arabic normally it triggers thinking it is the wake word, and so i thought i'll collect the data myself, and i did for both positive and negative. the thing is that i don't know how to let the openwakeword project use my data instead. i searched in the yaml file for something i can change in it but i didn't know what to change. if someone knows how to setup openwakeword to learn from my dataset please tell me the steps.