Spectral Editing: Repairing Audio in the Frequency Domain

Most explanations of this start with a definition and stop before they get to the part that affects your session.

This covers spectral editing frequency domain repair enhancement the way it actually works in a session: the mechanism, the numbers that matter, and the decisions that follow from them.

Nothing below is theory for its own sake. Every number here is one you can check on a meter, and every step is one you can run on the next track you open.

Key takeaways

Restoration removes what should not be there, it cannot add what was never captured.
Order matters: de-click, de-hum, broadband noise, de-reverb, then tonal repair.
Two gentle passes beat one aggressive pass.
Always compare level-matched, because quieter reads as cleaner.

What restoration can and cannot recover

Restoration removes what should not be there. It cannot invent what was never captured. That single distinction explains most disappointing results.

Problem	Recoverable?	Method
Broadband hiss	Yes	Spectral noise reduction with a clean noise profile
Mains hum at 50 or 60 Hz	Yes, cleanly	Narrow notches at the fundamental and its harmonics
Clicks and crackle	Yes	Interpolation across the damaged samples
Digital clipping	Partly	Waveform reconstruction of the flattened peaks
Excessive room reverb	Partly	De-reverberation, at the cost of some transient detail
Missing high frequencies	No	If it was never recorded, there is nothing to restore

Restoration order. Getting this wrong makes every later stage worse.

Work in the right order

De-click and de-crackle first. Impulsive noise confuses every stage that follows.
De-hum next, with narrow notches. A wide cut takes musical content with it.
Broadband noise reduction third, using a profile captured from a genuinely silent passage.
De-reverb only if it is actually a problem, and always less than you think you want.
Tonal repair last, once the noise is gone and you can hear what is really there.

Every restoration stage is destructive. Two passes of gentle reduction almost always beat one aggressive pass, because the artefacts do not compound the same way.

The artefacts to listen for

Musical noise: a watery, chirping quality in the top end from too much spectral reduction.
Dulling: the sense that the life has gone with the hiss, which means the threshold is too aggressive.
Pumping: the noise floor breathing in the gaps between phrases.
Transient softening: from de-reverb or de-click set too hard.

Bypass often, and always compare at matched level. Noise reduction lowers overall level, and quieter reliably reads as cleaner even when it is not.

Salvaging a clipped or damaged master

Clipping flattens the tops of waveforms. Declipping estimates the curve that should have been there and rebuilds it, which works well for brief overs and poorly for a file that is clipped throughout.

Reduce the file level by 3 to 6 dB first so the reconstruction has somewhere to go.
Run declipping before any other processing.
Check the reconstructed peaks in a waveform view, not just by ear.
Follow with gentle limiting, not more clipping.

How it translates on real systems

Almost nobody hears your record on the system you made it on. The point of every decision above is that it survives the journey.

System	What it exposes	What to check
Phone speaker	No low end at all below roughly 500 Hz	Does the bass line still read from its harmonics?
Earbuds	Exaggerated width and sub, hyped top	Is anything sibilant or fatiguing?
Laptop	Thin midrange-only playback	Is the vocal still intelligible?
Car	Boomy low end and road noise	Does the low end turn to mush at 80 to 120 Hz?
Club system	Everything below 40 Hz, loud	Is the sub mono and controlled?

If it works on a phone and in a car, it works almost everywhere else. Those two are the honest tests.

Knowing when to stop

There is a point in every session where further work stops improving the record and starts merely changing it. Recognising that point is most of the skill.

You are undoing changes as often as you are making them.
You cannot articulate what the last move fixed.
The reference no longer sounds obviously better in any specific way.
You are adjusting things you would not notice as a listener.

Bounce it, sleep on it, and listen once in the morning on a system you did not use to make it. That single listen tells you more than another two hours ever will.

Frequently asked questions

What is the most important thing to get right with spectral editing frequency domain repair enhancement?

Tonal balance and translation. A track that holds up on a phone, in a car and on monitors is finished. One that only works on your monitors is not.

What loudness should I target?

-14 LUFS integrated with a -1 dBTP ceiling covers almost all streaming. Go quieter for acoustic and classical material, louder only where the genre genuinely demands it.

How much headroom should the mix have?

Three to six decibels, with no limiter on the mix bus. That gives the mastering stage something to work with.

How do I know when it is finished?

When it holds up level-matched against a commercial reference, translates on several systems, keeps its low end in mono, and your changes have stopped improving it.