Distortion in audio: what causes it and how to fix it

Distortion is any unintended change to the shape of an audio waveform between the sound happening and you hearing it. If what comes out is not a faithful scaled copy of what went in, the signal has been distorted.

That is a broad definition, which is why the word covers so many different complaints. A crackling voice note, a muddy conference call, a buzzing guitar amplifier and a fuzzy laptop speaker are all distortion, and they have almost nothing in common except the symptom.

So the useful question is never how to remove distortion in general. It is which kind you have, and at which point it entered the recording.

By far the most frequent cause is feeding a stage more signal than it can carry. In the digital world this produces clipping, where peaks are cut off flat against a hard ceiling.

The identifying signs are consistent. It appears only on the loudest moments and disappears when the speaker is quiet. It sounds crunchy or gritty rather than muffled. And zooming into the waveform shows flat plateaus where there should be peaks.

If those three things are true, you have an overload problem, and the fix is a gain-staging fix rather than a processing one.

Failing hardware. A blown speaker cone or a damaged microphone capsule distorts at every level, not only on peaks. Test by playing a recording you know is clean through the same speaker, and by recording the same voice through a different microphone.

Cables and connections. A partly unseated jack or a damaged cable produces intermittent crackling that changes when you move the cable. This is the easiest cause to rule in or out and the one people check last.

Heavy lossy encoding. Compressing a file to a very low bitrate introduces artefacts that sound watery or metallic on sustained sounds, especially on music and on sibilance. This distortion lives in the file, so re-exporting from the original at a higher bitrate is the only cure.

Processing pushed too far. An aggressive noise reducer, a heavy equaliser boost or a stack of compressors can each generate distortion that was not in the original recording. If your raw file sounds fine and your export does not, work backwards through the chain.

Play the file on a second set of headphones or speakers. If the distortion disappears, the problem is your playback equipment, not the recording, and you can stop here.

Open the original, unprocessed recording rather than the export. If the raw file is clean, something in your edit or export introduced the problem.

Check whether the distortion tracks loudness. Present only on loud moments means overload. Present evenly throughout, including under quiet speech, points at hardware, cabling or encoding.

Zoom into a distorted moment in the waveform. Flat tops confirm clipping. A waveform that looks normal but sounds wrong points at encoding artefacts or playback equipment.

Start by lowering the gain on the distorted track rather than raising anything else. This will not undo distortion already recorded into the file, but it stops later stages compounding it and tells you how much of the problem is happening inside your project.

Remove processing one stage at a time and listen after each removal. It is common to discover the distortion was created by an effect rather than captured by the microphone, and no repair tool is needed at all.

If the distortion is genuinely in the source recording and it is mild, an audio enhancer can smooth the roughest moments enough to make an interview usable. Expect improvement rather than restoration: information destroyed at the input cannot be recovered.

For badly distorted material, consider whether the words matter more than the sound. A transcript, or a re-recorded narration of the same content, will often serve your listener better than a heavily processed original.

Set gain so your loudest expected moment peaks well below the ceiling, around -12 to -6 dBFS. Test with the loudest thing your speaker will actually do rather than their speaking voice at rest.

Record a short test and listen back on headphones before committing to a long session. Thirty seconds of checking prevents an hour of unusable audio, and it is the step almost everyone skips.

Export at a sensible bitrate. For spoken word, encoding well above the point where artefacts appear costs very little in file size and removes an entire category of complaint.