Noise Reduction

Most tools profile the noise from 2 seconds of silence, then subtract that fingerprint across the whole recording. Newer ones use trained models that separate speech from everything else. Both handle constant noise well and struggle with anything that shifts while you record.

Push either too far and you get the artefact everybody recognises: a swirling metallic edge on the voice, and words that vanish into the gaps between them. Use the smallest amount that makes a recording acceptable, then leave it there. Fixing the room, moving the microphone closer or turning the fan off beats every plugin, since processing can only remove what you already captured.