Is it possible to augment BirdNET's noise/background classes with custom samples?

[tphakala]

I've been training my own classifier and have collected numerous audio clips of wind, human-made noise, background hiss, and other environmental sounds to use as background class samples for my model. However, it seems these non-event classes don't affect the main BirdNET model weights at all. Is there a way to augment BirdNET's noise/background classes with my own samples to potentially reduce false positives?