FR: Add ability to specify multiple target words + negative words to automatic training colab
twitchyliquid64 opened this issue · comments
Heya!
I'm working on training a bunch of keywords, and it would be useful if I could add multiple pronunciations (i.e I'm australian but my gf is american) as well as negative examples (training 'turn on' vs 'turn off').
The config file used in train.py does support some features that might be similar to what you are describing. The more complete example training notebook shows how you can adjust the config parameters for greater control over the training process.
Specifically:
target_phrase allows you specify multiple wakewords (including those with different pronunciation) when training a model.
custom_negative_phrases allows you to specify specific words/phrases that will be added to the negative training data so (hopefully) the model will be less likely to activate on them.