Remove Mouth Clicks from Audio: The Manual Edit That Works

Why clicks survive every denoiser

A mouth click is a tiny acoustic event: a few milliseconds of broadband energy produced when the tongue, palate, or lips part in a particular way. It is over before a reduction filter has time to analyze it, and it does not resemble the steady noise bed that filters are built to learn. The click passes through a denoise pass essentially untouched, which is why creators who run their recording through every tool they own still hear clicks at the end.

The same property that defeats filters makes clicks easy to see. On a waveform, a click is a spike that stands out from the speech around it, and on a spectrogram, it is a short vertical streak. Locate them visually and the edit becomes mechanical rather than mysterious.

Set the right expectation before starting: a recording with twenty clicks needs twenty small edits, and that is not a failure of the tools. It is the nature of impulse noise, and the manual pass is how professionals handle it too.

Finding the clicks without losing an hour

The difference between a twenty-minute click cleanup and a two-hour one is the search strategy. Listening at normal speed catches the loud ones and misses the rest; scanning visually catches everything, but only if the view is readable.

1. Turn off or bypass any noise reduction first, so the waveform shows the recording as captured.
2. Zoom so that one second fills several inches of screen; clicks resolve into clear spikes at this scale.
3. Play a short loop with the waveform visible and mark every spike you hear, rather than deciding while listening.
4. Work in passes of a few minutes; freshness matters more than speed for this kind of detail work.
5. Keep a list of timestamps as you go, so the edit pass can proceed without re-hunting.

The manual edit, three ways

Once a click is selected, three techniques cover almost every case:

• Reduce the selection: drop its gain by several decibels so the click softens rather than disappearing; this keeps the timeline intact and sounds least invasive.
• Replace with room tone: copy a moment of clean background from the same recording and paste it over the click, which works best when the click fell in a pause.
• Soften or repair with a tool: some editors ship click-repair or spectral tools that smooth the spike; useful for stubborn cases, with the same restraint as any processing.

Reducing instead of deleting

The instinct to delete the click's region is understandable and usually wrong. A deletion shortens the audio by a few milliseconds, which can shift timing against a video or create a subtle hole in the room tone. Reducing keeps the duration intact and leaves a quieter click that usually passes unnoticed.

The amount is worth calibrating per click. Louder clicks tolerate deeper reduction; a faint click needs only a nudge, and over-reducing the surrounding speech becomes audible as a small dip. When in doubt, reduce less and listen in context.

Room-tone replacement that sounds invisible

Pause clicks are the easiest wins, because a pause contains nothing but room tone. Collect a few seconds of clean background from the same session, and paste a matching snippet over each offending click with short crossfades at both edges. Match the level as well as the content; a replacement that is slightly louder or quieter than its surroundings is more noticeable than the original click.

The technique is worth practicing on a throwaway file once. When the paste, the level, and the fades are right, the edit is genuinely invisible, and the file keeps its natural pauses instead of becoming a series of gaps.

The cleanup pass around the clicks

Clicks live on top of everything else, including the noise bed, and lowering that bed makes the manual pass easier. With the background quieter, each click is more obvious visually and each softened click blends more convincingly, because the contrast between the click and its surroundings has shrunk.

The order that serves most recordings: run the bed through a cleanup pass such as SimpleClean first, do the manual click edits on that cleaned copy, and keep the original untouched for reference. If the cleanup changes the character of the room in a way you dislike, the click edit can be redone from the raw source without starting over.

Prevention: hydration, angle, and pace

Most clicks are chemistry and geometry, which means most of them are preventable:

• Drink water before the session, and keep a glass nearby; a dry mouth clicks more.
• Avoid dairy and very sugary drinks immediately before recording if you know they affect you.
• Speak slightly across the microphone rather than directly into the grille, keeping lips out of the direct path.
• Take a sip and a pause between paragraphs; rushed speech produces more clicks than measured delivery.
• Keep the take rolling through a click-heavy stretch and edit it afterward; stopping to re-record each click costs more than the edits.

A five-click test before bulk editing

Before working through a hundred clicks, calibrate on five. Open a representative stretch of the recording, fix the first five clicks with the techniques above, and audition the fixes against the raw take. Five edits answer the questions that matter: how deep a reduction this recording tolerates, whether room-tone replacement is available at the right level, and how long each fix takes at this clipboard speed.

The calibration usually pays for itself twice. It prevents the classic bulk-edit trap, where forty edits are done one way and then redone because the approach was wrong, and it gives an honest estimate of the session's length before committing the evening to it.

What automatic de-click tools bring

Restoration suites include automated de-click and de-crackle processes that scan a file and repair impulses without manual selection, and they have a real place in this workflow. Against dense click fields, dozens within a minute, the automatic pass restores the working speed that manual editing cannot, and its settings usually expose a sensitivity and an amount to tune against a sample.

Their limits match their design. An automatic clicker cannot tell a mouth click from a stick tap from a consonant, so it errs on the side of caution and leaves some events while occasionally dulling sharp syllables. The division that works: automatic pass first for the dense cases, manual pass after for everything the scan left behind, and a comparison against the original at every step.

When to leave clicks alone

Not every click needs removing. Audiences tolerate occasional mouth noise, and an edit that introduces a dip or a jump is worse than the click it removed. A practical threshold: leave clicks that fall under words, and prioritize the isolated ones in pauses, which are the ones listeners actually notice. The five-click test from earlier doubles as a tolerance check: if even the well-edited clicks seem like an improvement worth keeping, continue; if they create more distraction than they remove, the raw take was telling you something.

Related guides

Keyboard clicks follow the same impulse logic from a different source, covered in removing keyboard noise from audio. If a mouth-noise cleanup has already been over-processed into a metallic texture, the recovery walkthrough is the way back, and cleaning podcast audio online covers the publishing workflow around the edit.