Test my model recognition is low with precise-train，and It would false recongition.

[molyswu]

Hi, How many examples model(hey-mycroft.pb or hey-mycroft.net)do you have in your precise-data-models? How many negative samples (not-wake-word)and how many positive samples(wake-word)?I training models and test for false recongition。The sample(wake-word) is 500M, the negative samples (not-wake-word) is your pdsounds_march2009.7z.

Thank you for your help

[MatthewScholefield]

The hey-mycroft model uses around 80k samples to train, however that many is certainly not necessary. I've been meaning to update the tutorial. The pdsounds dataset isn't actually the best negative sample set.

Try using the speech commands dataset(latest download here). It's an archive with a series of folders with a bunch of samples of different words. You can just drop that in the not-wake-word folder.

Also, to test, you can move one of the speech-command words into the wake-word folder to train a model using only their data and see if it works (this would make sure there is nothing different about your microphone that Precise is messing up on).

Hope this helps.

[Chen1399]

Dear sir, Should I add natural noise in not-wake-word folder ? I find my train result will be waked up by natural or speech.

[lyuzhilinag]

Dear sir, Should I add natural noise in not-wake-word folder ? I find my train result will be waked up by natural or speech.

I believe repeating 'Reducing False Activations' in https://github.com/MycroftAI/mycroft-precise/wiki/Training-your-own-wake-word#how-to-train-your-own-wake-word would help.