What is the principle of mycroft-precise model training

I used mycroft-precise to train a new model and found that audio without wake-up words would also be predicted with high scores.Looking at the source code, I found that the wake word data will set label 1 when training the model, and the data without wake word set label 0, but when verifying the model, I can’t get the label value and only get the score. What is going on?