Can't train the model

[zxcqirara]

I have recorded 250 WAV files with my wakeword and 1800 files of just ambient sound. I tried to train a model (with parameters from the rustpotter-cli docs) and used 100 test wavs with keyword and 100 with ambient noise + some voice (without keyword) + keyboard typing, etc., while I was testing it, it made false positives when I just was making sounds like "uuugh". Then I decided to record [none] test wavs again, so that they were just with ambient noise, without anything else. Then while testing it just almost didn't recognize my voice at all. One more odd thing is that it was showing exactly 50% of accuracy while training