Over-processing
Also known as: over-cleaning
Over-processing is cleaning a recording so aggressively that the voice itself degrades: word endings vanish, consonants soften, and artifacts creep in. The noise is gone, but so is some of the speech. The fix is to clean gently and measure what was removed.
Why it matters
Over-processing is the invisible failure of audio cleanup. It looks like success, the hiss is gone, so it is easy to ship. But the same aggressive setting that removed the noise also filed off the quiet edges of speech, and the result sounds underwater or lispy in a way that is hard to trace back to the cause.
The instinct that causes it is treating more cleanup as always better. It is not. Past a point, every extra decibel of noise removed costs speech.
In practice
Clean in small steps and stop early. Leave a little noise rather than chasing silence, compare against the original constantly, and if you can, transcribe both versions: a word error rate that rises after cleanup is over-processing showing up as a number.