In this lesson
A clipped take stays clipped after the fader
Clipping at the preamp or converter is written into the file. The fader only scales playback, so the distortion gets quieter and stays.
· 8 min read · Lesson 4 of 12 in Vocal Production
In short
- Set input gain on the loudest section of the song, sung for real, because anything that clips at the preamp or converter is written into the file.
- Fader, clip gain and normalizing multiply every sample by the same number, so they turn the distortion down by exactly as much as the vocal.
- When a take clips, lower the gain upstream and record again, and save the declipper for a few words on a take you cannot replace.
In this lesson (8)
The chorus that crackles
The verses went down clean. In the last chorus the singer opened up, the interface's clip light blinked on two words, and nobody stopped the take because it felt like the one. In the mix those two words have a fizzy edge on the vowel. You pull the vocal fader down 6 dB. The edge gets quieter along with everything else and stays exactly as rough.
The converter flattened those peaks before the take reached the disk, so every later move, the fader included, works on a wave whose tops are already gone.
Why it matters
The file holds what the converter saw
In a typical recording chain the microphone feeds a preamp, the preamp feeds the analog-to-digital converter, and the converter's numbers are written to the file. The channel's inserts and fader come after that, on playback, which is why recording level is set at the preamp and not at the fader (Nichols, 2018). Moving the fader during a take changes what you hear and leaves the file alone.
A converter with fixed-point output has a largest number it can report: full scale, or 0 . When the voltage from the preamp asks for more, every sample past that point is written as the maximum value and the top of the wave comes out flat (Pohlmann, 2011). The preamp has a ceiling of its own, its maximum output voltage, and a preamp driven past it flattens the wave before the converter ever sees it. Either way the damage is in the recording, and the DAW meter only reports it afterwards.
That is also why a session does not rescue the take. A typical interface converter delivers fixed-point samples, so a clipped sample arrives as full scale, and float stores that full-scale value perfectly. Float headroom protects you between plugins inside the DAW. It starts after the converter, which is where this problem began. Field recorders that merge two converters into one float file are the exception, which the float lesson also covers.
The science
A fader is one multiplication
Turning a fader down by dB multiplies every sample by the same number:
For 6 dB, . Every sample is halved, the flat ones included. A flat top at full scale becomes a flat top near -6 dBFS with exactly the same shape. A peak meter now shows no at all, which is how a turned-down clipped take slips through a session unnoticed. Gain cannot tell which samples were clipped, and even if it could, the true values are gone: the file only says they were at least full scale.
The shape is what you hear. A flat top has sharp corners the original wave never had, and corners mean new frequencies. Clipping that treats both halves of the wave the same way adds odd , the same ones a clipper plugin adds on purpose. Work it through for a sine driven 3 dB past full scale and the third harmonic comes out 17.8 dB below the note. A voice is not a sine, so its exact numbers differ, but the next step does not depend on them.
Now pull that clipped note down 6 dB. The note drops 6 dB, the harmonic drops 6 dB, and the gap between them, which is how much distortion there is, stays at 17.8 dB. The same holds for and for normalizing, because each of them is one multiplication.
In the demo, listen to the edge on the loudest notes, not the overall level. When the clipped phrase is turned down, ask whether the rough edge went away or only got quieter.
Listen
Turn down a clipped take
A phrase recorded too hot clips at the converter. Pull the fader down afterwards and the level drops, but the flattened peaks and their distortion stay.
Loading demo…
Start with your speakers or headphones low.
Repair tools exist. A declipper looks at the samples that did not clip and estimates what the flattened peaks would have been, using assumptions about how audio usually behaves, such as a sparse spectrum. Záviška and colleagues (2021) reviewed the popular methods and tested them on real audio. Because the estimate is built from the samples that survived, the more of a phrase is clipped, the less a declipper has to go on. A declipper can rescue a few clipped syllables. What it gives you is an estimate, so a retake is still the better fix while the singer is in the room.
Try it
Clip a take on purpose
0 of 7 steps done
Fader, clip gain and declipper all change the clipped take after the fact. The clean take is clean because its gain was set before the converter.
Common mistake
Setting the gain on a polite run-through
The usual cause is a level set while the singer marks the part. They sing softly for the soundcheck, then give the real take everything they have. A singer who cannot hear themselves in the cue pushes even harder, which is one more reason to get the headphone balance right first. Set the gain while they sing the loudest section of the song at full performance level, and check the peaks again after the first real take.
The opposite mistake is treating 0 dBFS as a target, as if a hot take were a better take. At 24-bit the format's floor sits about 144 dB below full scale, and the noise of the mic, the preamp and the room is far louder than that, so leaving clear room between your peaks and full scale costs you almost nothing. Peaks around -18 to -12 dBFS are a common starting point, with more room for a dynamic singer.
Takeaway
Fix it where it happened
When a take clips, go upstream: lower the preamp gain, switch in a pad if the mic or interface has one, or ask the singer to step back a little. Then record again. Check the loudest moments of each take as you go, because the clip light only flashes for a moment while the file keeps the flat tops. If the clipped take has the best performance, comp the damaged words from another take or try a declipper on just those words before you give up on it.
dBFS
Decibels relative to full scale, the largest value a fixed-point file or converter can hold, so levels below it are negative numbers. Inside a floating-point mix a signal can read above 0 dBFS and stay intact; it clips where it leaves float or passes through a plugin that clips. dBTP is the same scale measured as true peak, and it can read above 0.
32-bit float
A sample format that stores a scale factor with every value, so levels above 0 dBFS inside the DAW are kept instead of clipped. Overs still clip at a fixed-point export, at the interface output and inside any plugin that clips.
Headroom
The space between the loudest peak in a signal and the level where it would clip.
Clipping
Cutting off the top of a waveform at a fixed level. Hard clipping flattens the peaks; soft clipping rounds them, which adds fewer harsh harmonics.
Harmonics
Frequencies above a note's fundamental. In pitched sounds they sit at whole-number multiples of it, and their balance is a large part of timbre.
Clip gain
A gain change applied to a region of audio itself. In most DAWs it acts before the plugins on the track, so unlike the fader it changes what they hear.
3 questions
Check yourself
Sources
Show all 3Hide sources
- Nichols, P. (2018, January 7). Downloadable charts to understand audio signal flow in a DAW. iZotope. www.izotope.com/community/blog/understanding-audio-signal-flow-in-a-daw
- Pohlmann, K. C. (2011). Principles of Digital Audio (6th ed.). McGraw-Hill.
- Záviška, P., Rajmic, P., Ozerov, A., & Rencker, L. (2021). A survey and an extensive evaluation of popular audio declipping methods. IEEE Journal of Selected Topics in Signal Processing, 15(1), 5-24.
Next lesson · 5 of 12
How proximity effect turns warmth into mudSinging close to a directional mic adds bass that grows as the distance shrinks. A high-pass is the wrong shape to undo it, so set the distance at the source.
Written by Virzy Guns
Songwriter and producer, founder of Virzy Guns Production. Now building HealingWave.