Mouth Clicks in Audiobook Narration: What Software Can Fix and What Still Needs Editing

A recording can have a good microphone, a quiet room and perfectly acceptable levels, but a few wet clicks between words can still make the narration feel rough.

And on an audiobook, you are not dealing with thirty seconds of audio.

You might have ten hours of narration.

That changes the problem completely.

The question is not really:

“Can I remove a mouth click?”

Of course you can.

The more useful question is:

“How much of this can I fix automatically without damaging the voice?”

That is where de-click software helps, but also where it can go wrong.

First, what actually counts as a mouth click?

Mouth clicks are short noises created by movement inside the mouth.

They can sound like:

• small wet clicks
• lip smacks
• saliva noises
• little crackles between syllables
• sharp noises just before or after a word

Some are obvious.

Others sit inside consonants and are much harder to separate cleanly.

That distinction matters because software does not understand a click in the way you do.

It sees a very short transient in the waveform and tries to decide whether that transient belongs to the performance or not.

Sometimes that is easy.

Sometimes the “click” is sitting right beside a T, K, P or another sharp consonant.

Remove too much and you can start removing part of the word.

That is one reason aggressive de-click processing can make narration sound soft, lisped or slightly unnatural.

Can you stop mouth clicks while recording?

Sometimes you can reduce them.

But I would be wary of treating recording technique as a complete solution.

Things like hydration, microphone position and taking breaks can help some narrators.

If your mouth is extremely dry, that can obviously make the problem worse.

But narrators can do everything “right” and still produce clicks.

Some voices and speaking styles simply create more of them than others.

And once you are recording for hours at a time, the occasional click is almost inevitable.

So I would not build an audiobook workflow around the idea that you should somehow record a perfectly click-free performance.

A better workflow is:

reduce the problem at source where you can, then clean up what is left.

Why audiobook mouth clicks are harder than a normal voiceover

A thirty-second commercial gives you options.

You can zoom in.

Find the click.

Redraw it.

Use spectral repair.

Replace a syllable.

Do it again if necessary.

On a ten-hour audiobook, doing that to every small click can become a full-time job.

That is where automated de-clicking becomes useful.

If software can remove or reduce the obvious routine clicks across the whole recording, you can spend your time on the difficult ones instead.

That is a much better use of automation.

The aim is not:

remove every transient that looks suspicious.

It is:

remove enough of the repetitive problem that manual editing becomes manageable.

That difference is important.

What de-click software is actually doing

A de-clicker is looking for very short transient noises that do not behave like the surrounding audio.

Depending on the tool, it may:

• detect the transient
• remove or attenuate it
• reconstruct the missing audio from the surrounding waveform
• apply processing only where it thinks a click exists

That works well when the click is isolated.

For example, a wet click in a quiet gap between two words is usually much easier to treat than a click sitting directly inside a hard consonant.

The closer the unwanted noise is to useful speech, the harder the decision becomes.

And automated tools do occasionally make the wrong decision.

The consonant problem

This is probably the biggest thing I would listen for when testing any de-clicker.

A click and a consonant can look surprisingly similar.

Both can contain a sharp, fast transient.

If you make a de-clicker too sensitive, it can start treating bits of actual speech as unwanted noise.

You may notice:

• softened T sounds
• dull K sounds
• less definition at the start of words
• consonants that feel smeared
• words that sound slightly “processed”

The click might technically be gone.

But you have also removed some of the narrator.

That is not a good trade.

For audiobook work, I would always prefer a few difficult clicks that need manual attention over a whole chapter whose consonants have quietly been damaged.

So should you de-click the whole chapter?

Usually, yes, but conservatively.

For long-form narration, a sensible approach is to use automated de-clicking as a first pass.

Run it at a setting that catches the obvious routine noises without chewing into the speech.

Then listen.

Do not assume the strongest setting is the best one.

I would specifically check:

• hard consonants
• whispered passages
• very close-mic sections
• emotional dialogue
• sections with lots of saliva noise
• quiet pauses

If the voice still sounds like the original performance and most of the routine clicks are gone, the processing is doing its job.

If the chapter suddenly sounds smoother but slightly blurry, you have probably pushed it too far.

When manual editing still wins

There are some clicks I would not trust a global automated setting to solve.

The main ones are severe clicks sitting inside speech.

If a loud click overlaps the start of an important consonant, software has very little clean information to work with.

That is when manual repair may be safer.

Depending on the audio, that could mean:

• attenuating the click
• using spectral repair
• redrawing a tiny waveform section
• copying a clean piece of nearby room tone
• replacing the word or phrase from another take
• re-recording the line

That sounds slower because it is slower.

But you are only doing it on the difficult cases.

That is the whole point of the automated first pass.

Don't confuse mouth clicks with every other small noise

Not every short noise in narration is a mouth click.

You can also have:

• plosives
• sibilance
• microphone bumps
• clothing noise
• headphone cable noise
• tongue noise
• breaths
• background ticks or electrical noise

Different problems need different treatment.

For example, a plosive is not something I would try to solve with a de-clicker.

Likewise, aggressive processing on breaths can make narration feel unnatural very quickly.

If harsh S sounds or plosives are the real problem, read our guide to de-essing and plosives.

And if you are trying to decide what to do with breaths, I have a separate guide on breaths in audiobook narration.

The bigger danger is over-cleaning

Once you start noticing clicks, it is very easy to become obsessive.

You zoom further in.

Find another tiny noise.

Remove it.

Find another.

Remove that too.

Eventually, you can end up cleaning things that a listener would never have noticed.

This is one of the weird things about audio editing.

Once you know where a problem is, you hear it every time.

The listener does not.

They hear the chapter once, in context.

That does not mean you should leave obvious mouth noise everywhere.

It means the standard should be distracting, not microscopically imperfect.

If the narration sounds natural and nothing is pulling your attention away from the story, you are probably in a good place.

I would much rather hear a tiny natural mouth sound occasionally than narration that has been processed so aggressively that the voice feels artificial.

That is the same principle behind not using too much noise reduction.

Where mouth-click cleanup should sit in the chain

I would generally deal with obvious cleanup before the final mastering stage.

That means:

1. edit the performance
2. choose your takes
3. remove major mistakes
4. deal with technical cleanup such as clicks and noise
5. control tonal balance and dynamics
6. master towards the required delivery range
7. check the final exported file

There is a practical reason for that.

Mastering raises, compresses and controls the recording.

If you leave a click in and then bring the whole chapter up in level, you may make that click more obvious.

Deal with the problem before the final level stage where possible.

What about ACX?

Mouth clicks are not really an RMS or peak problem.

You can have a file that meets the technical loudness requirements perfectly and still contains distracting mouth noise.

That is why passing a meter is not the same thing as having a finished audiobook.

Human listening still matters.

ACX's technical measurements tell you whether the file sits inside the required delivery window.

They do not tell you whether the chapter is pleasant to listen to.

If you are having problems with the actual measurements, read why ACX rejected my audiobook audio.

Mouth clicks belong to a different part of the quality-control process.

What I would automate

For a long audiobook, I would absolutely automate the routine part.

Running hours of narration through a sensible de-click pass makes far more sense than hunting every small transient manually.

But I would keep the difficult decisions human.

That means:

Automate:

• obvious routine mouth clicks
• repeated low-level saliva noise
• first-pass cleanup across long chapter files

Check manually:

• clicks inside consonants
• loud isolated mouth noises
• anything the de-clicker has damaged
• emotional or unusual speech
• obvious remaining distractions

That is the balance I tend to prefer for spoken-word processing generally.

If the job is repetitive and technical, automation is useful.

If fixing it means deciding what part of the performance is important, I want a person involved.

That is also the principle behind what audiobook narrators should automate and what should stay manual.

How Optivox handles it

Optivox includes mouth-click reduction as part of the audiobook cleanup pass.

The aim is to reduce routine clicks without treating every sharp transient as a mistake.

But it is not magic.

A severe click sitting inside an important consonant can still need manual editing.

And if a recording contains a lot of extremely loud mouth noise, it may be better to address the recording technique as well rather than asking processing to do all the work.

That is an important limitation to be honest about.

The useful part is removing a lot of the repetitive cleanup before you start listening for the exceptions.

Listen like a listener

When you finish the chapter, stop staring at the waveform.

Put headphones on.

Let it play.

If you are listening for every microscopic click, you will always find something.

Instead ask:

Is anything pulling my attention away from the narrator?

That is the standard that matters.

A good cleanup pass should not make you think:

wow, they really removed every mouth noise.

It should make you stop thinking about the audio at all.

You hear the narrator.

You hear the story.

And the technical work disappears into the background.