What Clipping Actually Is (and Why It Sounds the Way It Does)
You've heard it: a voice recording or music track that sounds harsh, buzzy, almost "crunchy" at its loudest moments. The vocal or instrument sounds fine at lower levels but distorts badly whenever it gets loud. That's clipping.
Here's what's happening at the signal level.
Every audio recording system — analog or digital — has a maximum level it can handle. In digital audio, this ceiling is called 0 dBFS (decibels relative to full scale). When the input signal exceeds 0 dBFS, the system can no longer represent the waveform accurately. In digital recording, the values simply max out at the ceiling and stay there. On a waveform display, you can see this as a flat top on the peaks — the rounded peaks of a normal audio waveform are cut off and replaced with a flat line.
Mathematically, what happened is this: a smooth sine wave with a peak at +3 dBFS got recorded as a sine wave with the top 3 dB sliced off and replaced with a rectangular plateau. When you play this back, the abrupt transitions from the sloped part of the wave to the flat ceiling contain high-frequency energy that wasn't in the original signal. This is the harsh "buzz" you hear.
The severity of clipping depends on:
- How far over the ceiling the signal went — 0.5 dB of clipping sounds very different from 6 dB of clipping
- For how long — a single 1-ms peak clipping is often inaudible; continuous clipping on sustained sounds (vocals, synths) is severe
- The nature of the signal — clipped percussion is often less audible than clipped vocals because the transient nature of drums means the flat sections are very brief
Analog Clipping vs. Digital Clipping
These are often confused but they behave differently.
Digital hard clipping is the scenario described above — the waveform literally cannot be represented above the ceiling and gets truncated. The result is "harsh" and "buzzy" because the waveform edges become essentially vertical.
Analog soft clipping occurs in tube amplifiers, tape saturation, and certain analog circuit designs. The electronics gradually compress the signal as it approaches the system's limits rather than cutting it off abruptly. The peaks still get flattened, but the transition is smoother, which means the high-frequency content generated is less harsh. This is why a recording that was "driven" through a warm analog tape machine sounds different from the same signal clipping a digital input — the tape clips "softly" and the digital input clips "hard."
"Overdriven" recording chains — where multiple stages are each adding a little distortion — produce complex, layered clipping that's often harder to repair than simple hard digital clipping because the distortion signature is irregular.
How to Identify the Severity of Clipping
Before attempting any repair, you need to assess what you're dealing with.
Step 1: Visual inspection
Load the audio into any DAW (Audacity, Adobe Audition, Logic, Reaper) and zoom in on the waveform. Look for flat tops on peaks. If you see them, you have clipping.
Step 2: Quantify the clipping
Audacity has a "Show Clipping" function that highlights clipped sections in red. iZotope RX shows clipped regions in its waveform view. Count how many clipped sections there are and how long they are:
- Occasional isolated peaks (< 5 ms each): mild, likely repairable
- Frequent peaks (several per second): moderate, repair possible with visible artifacts
- Continuous clipping during loudest sections: severe, partial repair at best
Step 3: Listen critically
Play the clipped sections. Does the distortion appear only at the loudest moments (transients), or is it present throughout sustained sounds? Transient clipping is easier to mask; sustained clipping of voices or sustained instruments is the hardest category.
Declipping Algorithms: How Repair Actually Works
Declipping is technically difficult for a precise reason: the clipping process destroyed information. The original peak was, say, +3 dBFS — but all we have is the value at the ceiling. We don't know exactly what the waveform was doing above that ceiling. Repair algorithms have to infer what was likely there.
The most sophisticated declipping algorithms work by:
- Identifying the clipped regions — any sample held at the maximum value
- Analyzing the surrounding unclipped waveform — using the shape of the waveform just before and after each clipped section to predict the slope, amplitude, and curvature of the missing peaks
- Reconstructing the peaks — inserting a mathematically plausible waveform shape that connects smoothly to the surrounding audio and whose spectral content is consistent with the unclipped portions
The better the algorithm, the more of the surrounding context it uses and the more sophisticated its model of what the original waveform "should" look like. Modern machine learning-based declipping tools can produce very good results for mild to moderate clipping because they were trained on millions of examples of clipped and unclipped audio pairs.
What You Can Fix Yourself
Mild clipping (occasional transient peaks, < 2 dB over ceiling)
Audacity (free):
Audacity doesn't have a dedicated declipping tool, but for very mild cases:
- Apply a hard limiter with a ceiling of -0.3 dBFS and make-up gain of -1 to -2 dB — this won't repair the damage but will prevent further degradation from any re-processing
- Apply very gentle saturation with a softening curve to round the remaining edges
This is a workaround, not a real repair.
Adobe Podcast (free, speech only):
Adobe's speech enhancement model was trained on a wide variety of recording conditions and can mitigate mild clipping artifacts in voice recordings reasonably well. Worth trying on any speech recording.
iZotope RX Elements — Clip Gain:
The Elements tier includes basic declipping. Effective for mild cases with isolated transients. The full RX suite's declipping is significantly more powerful.
Moderate clipping in known material
If you have a reference recording of the same content (a second microphone, an analog backup, a broadcast copy), you can use spectral repair combined with the reference to reconstruct the clipped sections. This is technically complex but can yield excellent results.
What Cannot Be Fixed
Severe sustained clipping of voices
When a vocal performance is clipped throughout most of its duration, there is no reference left from which to infer the original waveform. The fundamental pitch is usually recoverable but the harmonic structure is permanently altered. Even the best professional tools will leave audible artifacts on severely clipped sustained vocals — typically a "plastic" or "overprocessed" quality.
Clipping combined with heavy compression
A recording that was aggressively compressed and clipped has two layers of irreversible dynamic damage. Declipping can address the waveform truncation, but the dynamic compression that preceded it cannot be undone.
Low-bitrate digital recording with clipping
A recording that was captured at 8-bit resolution (common on older dictaphones, voice recorders, early digital answering machines) and clipped has both quantization noise and clipping distortion. The information loss is compounded.
The Honest Assessment
The question "can you fix this?" depends entirely on the answer to: how much information was destroyed?
Clipping is one of the few audio problems where the damage is a direct function of how far over the ceiling the signal went. A recording that peaked at +0.3 dBFS (barely touching the ceiling for brief moments) is almost certainly fully recoverable. A recording that was running at +6 dBFS continuously has lost substantial signal information that cannot be reliably reconstructed.
The professional restoration workflow for clipped audio:
- Declipping pass — reconstruct peaks algorithmically
- Equalization — compensate for frequency-dependent distortion
- Harmonic analysis — identify and reduce intermodulation distortion products
- Spectral repair — manual editing of the worst artifacts
- Limiting — prevent any remaining hot peaks from re-clipping downstream
This takes time, professional tools, and trained ears. For a recording that matters — an interview, a performance, a family recording — it's worth doing right.
At WefixSound, we assess every submission before quoting. If your recording is too damaged to produce a result you'd be satisfied with, we'll tell you that before taking your money. If it's recoverable, we provide a free sample so you can hear the difference yourself before committing.