Peak normalization turns a whole clip up until its single loudest sample hits the target. If the take has one loud spike (a plosive, a bump of the mic stand, a cough), that spike reaches the target first and the rest of the performance stays quiet. Ovrtone's region Normalize, the one its engineer uses, sets a vocal's level by its singing peaks instead, so one spike can't decide it.
What should you do instead?
Find the spike: look for one peak far above the rest of the waveform.
Deal with it alone: trim it, fade it, or turn down just that region.
Then normalize, or better, set the level by loudness (LUFS) or by the typical peaks of the performance rather than its single loudest moment.
How does Ovrtone do it?
Ovrtone reports outlier spikes separately from the take's real level, so a single spike doesn't decide how loud the vocal is, and it won't push a spike over 0 dBFS unless you ask it to.
Ovrtone (ovrtone.app) is a DAW for macOS with an AI engineer built in: you make the music, it does the engineering, with your own plugins. macOS 15 or later, Apple silicon.
