How to Fix Clipped or Distorted Vocals (And When You Can't)
Recording · 7 min read
Clipping is the one vocal problem where the honest answer is often "re-record". A converter that hits 0 dBFS throws away the top of the waveform, and a declipper can only guess what was there. On short clips of a few milliseconds that guess is invisible in a mix. On a whole chorus sung into the red it is not, and no EQ or saturation brings it back; the best you can do is make the distortion sound deliberate. Here is how to tell which case you have in two minutes, how to hide what will not fix, and what to check before the next take.
Three outcomes, in order of likelihood
Fix · Hide · Re-record
Short flat tops: fixable · steady fuzz: hideable · sustained clip: re-cut
The 60-second answer
Zoom into the waveform on the loudest word. Flat tops of a few samples to a few milliseconds on a handful of peaks: run a declipper first in the chain, threshold just under the clip level, and expect it to disappear. Tops flat for most of a syllable, or a fuzzy edge across the whole chorus: the damage is continuous and a declipper will only soften it. Then your tools are a dynamic EQ cut of 3–5 dB around 2.5–4 kHz where the fizz lives, a wide de-esser, a low-pass at 12–14 kHz, and a saturator at 15–25% mix so what remains sounds like a decision. If the lead still sounds broken in the mix, re-record the clipped lines, and only those, with the preamp 10 dB lower.
What clipping actually does to the file
When a signal exceeds the converter's maximum, every sample above the ceiling is written at the ceiling, and the rounded top of the wave becomes a flat line. A flat-topped wave carries extra odd harmonics stacked all the way up the spectrum, which is why clipping sounds fizzy and bright rather than warm. Declippers find those flat runs and draw a curve through them that matches the slope on either side. The shorter the run, the better the guess. A clip of 10 samples has one plausible shape. A clip of 500 samples has thousands, and the tool picks one that sounds smoother but is not what the singer did.
One exception: if the red light came from a plugin or the DAW master and your session runs at 32-bit float, nothing was lost. Float does not clip inside the DAW. Pull the gain down before the stage that overloaded and the waveform is intact. Only the converter on the way in, and the export on the way out, destroy samples. Check the raw audio file, not the channel meter, before you assume damage.
Four kinds of distortion, and which one you have
Converter clipping shows as hard flat tops and a sharp, crackling edge on loud words; this is the declipper's territory. Preamp overload happens before the converter, so the file may peak at −6 dBFS and still sound crushed: the tops are rounded, the sound is thick and furry, and no declipper can see it because there is nothing flat to find. Mic overload is rare but happens when a singer screams at 5 cm; it sounds like preamp overload with added rattle. Plugin clipping inside a chain is a makeup gain or saturator pushed past 0 into a non-float stage, undone by lowering the gain. The first and last are fixable. The middle two are hide-or-re-record. The headroom before mixing article has the levels that prevent all four.
Fixing clipped vocals step by step
| # | The move |
|---|---|
| 01 |
Diagnose in the waveform, not the meter Zoom to sample level on the three loudest words. Flat tops: converter clip. Rounded but crushed: preamp. Count the affected phrases. < 10 clipped peaks: fix · whole sections: hide or re-cut |
| 02 |
Declip first, on the raw clip First plugin in the chain, or an offline render. Threshold just under the clip level so it only touches the flat runs. Audacity's Clip Fix is free; dedicated restoration suites do better on longer runs. Threshold −0.5 to −1 dB under the ceiling · makeup gain −6 dB |
| 03 |
Cut the fizz where it lives Sweep 2–5 kHz with a narrow boost on a clipped word; the distortion jumps out at one spot. Cut it dynamically so clean phrases keep their presence. Dyn EQ 2.5–4 kHz · −3 to −5 dB · Q 2 |
| 04 |
De-ess wide, low-pass gently Clipping harmonics pile up above 8 kHz. A broadband de-esser and a soft low-pass take the edge off without dulling the vocal. De-esser 5–10 kHz wide · LPF 12–14 kHz · 6 dB/oct |
| 05 |
Make the rest sound intentional Consistent tape or tube saturation across the whole vocal disguises uneven distortion as texture. Blend, do not drench. Saturation mix 15–25% · low drive |
| 06 |
Replace what will not hide Comp clean words from other takes or the double. Re-sing the rest at the same mic distance. Then compress as normal. Crossfades 5–15 ms · same mic, same distance, gain −10 dB |
The trick: steal the peaks from the double
Most sessions with a clipped lead also have a double, a harmony stack or an earlier take of the same line, sung softer, that did not clip. The clipped parts of a word are the loudest 20–50 ms of a few syllables, not the whole phrase. Cut just those peaks out of the lead, drop the same 20–50 ms from the clean take into the gap, and crossfade 5–10 ms each side. The ear forgives a tiny timbre change on a transient far more than it forgives fizz. If the double is slightly off in pitch or timing, nudge it; a 10 ms slice does not need to match for long. With no clean source, the comping guide covers building the line from whatever you have, and the free online Voice Recorder will capture a scratch of the lines to re-sing when you are away from the interface.
Chains with the reasons left in.
The Vocal Chain Bible is 88 real vocal chains from hit records, as the engineers described them — mic, preamp, every plugin and setting, in order, with a free-plugin version of every chain. Every chain comes with the reason behind each setting, so you know what to move for your voice.
Get the Vocal Chain Bible — $48 →Mistakes that kill it
- Declipping after compression. The compressor has reshaped the flat tops into something the declipper cannot recognize. Declip the raw file, first in the chain.
- Turning the clip down and calling it fixed. Lower level does not remove the harmonics. A clipped vocal at −20 dB is a quiet clipped vocal.
- EQing the fizz out with a static cut. A permanent −5 dB at 3 kHz dulls every clean phrase too. Use a dynamic band.
- Re-recording with the same gain. Drop the preamp 10 dB, aim for peaks at −12 to −6 dBFS, and check the loudest line before the full take. The Gain Staging Calculator gives you the target.
Keep going
FAQ
Can clipped vocals be fixed?
Short clips of a few milliseconds on a handful of peaks, yes, with a declipper run before any other processing. Long stretches of clipping or preamp overload can be softened and disguised but not restored. If the whole chorus clipped, re-record it.
What does a declipper actually do?
It finds runs of samples stuck at the ceiling and redraws a curve through them from the slope on either side. The guess is very good for short runs and increasingly wrong for long ones. It cannot see preamp or mic overload at all, because those leave no flat tops.
Why does my vocal sound distorted when the meter never hit red?
The overload happened before the converter, in the preamp or the mic, or the file came from a session that clipped internally at a non-float stage. The DAW meter only shows what arrived. Check the raw waveform for rounded, crushed peaks and check the interface gain.
Who's behind this
Ivan Meshcheriakov · IvanFromRnD
5+ years ago I moved to LA with zero connections and started building from nothing — 14-hour days, meeting people and interning at studios. It took me 5+ years to meet and interview 100+ Grammy-winning engineers — then I turned it all into the Vocal Chain Bible.
Read my full story →