Maybe the RMS is too low. Maybe your peaks are too high. Maybe the numbers appear fine but the audiobook still has a problem when it reaches quality review.
The important thing is not to start changing settings at random.
ACX failures generally fall into two categories: technical measurements that are outside the required range, or problems with the recording, editing or file itself. Those require different fixes.
ACX currently requires each submitted file to measure between -23 dB and -18 dB RMS, stay below -3 dB peak, and have a noise floor no higher than -60 dB RMS. Files also need to meet requirements around format, consistency, spacing and audio quality.
And one useful distinction: passing ACX Audio Lab does not mean every possible issue has been checked. Audio Lab analyses things such as RMS, peaks, bitrate, sample rate, channels and spacing, but ACX says it does not detect noise floor or editing errors.
So if your audiobook has failed, start by working out what kind of failure you actually have.
| Problem | Common cause | First thing to check |
|---|---|---|
| RMS too low | Narration is too quiet overall | Overall level and dynamics |
| RMS too high | Narration has been pushed too loud | Gain, compression and limiting |
| Peaks above -3 dB | Loud words or transients are escaping | Peak control / limiter |
| Noise floor above -60 dB | Room, microphone or equipment noise | Recording noise before increasing level |
| Audio sounds gated or unnatural | Noise reduction or gating is too aggressive | Room tone and transitions |
| Audio Lab fails format checks | Export settings are wrong | MP3, bitrate, sample rate and channels |
| Numbers pass but QA still finds problems | Clicks, plosives, outtakes, inconsistent sound or editorial errors | Listen to the finished file, not just the meters |
Your RMS is too low
This is one of the easiest problems to misunderstand.
If a chapter measures below -23 dB RMS, the obvious response is to turn the whole file up.
Sometimes that is all it needs.
But increasing the level raises everything, including the background noise.
A chapter sitting at -26 dB RMS with a borderline noise floor can end up inside the loudness target while pushing the room noise outside the acceptable range.
That is why loudness and noise floor should not be treated as completely separate problems.
If the recording itself is reasonably clean, the usual approach is to control the dynamics first, then bring the overall level towards the required range while keeping enough peak headroom.
If the recording is noisy, simply adding gain may fix one measurement and create another problem.
Your RMS is too high
ACX currently specifies an RMS range of -23 dB to -18 dB.
If your file is louder than that, first check whether the entire chapter has simply been mastered too hot or whether compression has reduced the dynamics so aggressively that almost everything is sitting at a similar level.
You generally do not need audiobook narration to be pinned against the top of the permitted range.
The goal is consistent, comfortable narration, not maximum loudness.
Reduce the overall level where necessary and listen again. If the voice suddenly feels more natural when you stop pushing it, that tells you something.
Your peaks are too high
A file can have perfectly reasonable average loudness while still containing individual peaks above ACX's limit.
ACX requires peaks to remain below -3 dB.
These usually come from isolated moments: an excited line, a hard consonant, a laugh, a sudden increase in projection or simply a section recorded slightly closer to the microphone.
Peak control is normally dealt with towards the end of the processing chain.
A limiter can catch occasional peaks without requiring you to turn the entire chapter down.
But if the limiter is constantly working hard, that usually points to another problem. The recording may be extremely inconsistent, the compression may need adjusting, or individual sections may need attention before mastering.
Your noise floor is too high
This is where trying to fix everything during mastering can become frustrating.
ACX specifies a noise floor no higher than -60 dB RMS.
If your room noise is already close to that threshold before you bring the narration up to the correct level, increasing the voice can also bring the noise up with it.
Common sources include computer fans, air conditioning, traffic, electrical hum, microphone self-noise and the general sound of the recording space.
Moderate noise can often be reduced successfully in post-production.
But there is a limit.
If heavy noise reduction is required just to force a recording into specification, you can start introducing watery artefacts, dull consonants or an obviously processed voice.
At that point the question stops being “Can I make the meter pass?” and becomes “Does this still sound like a professional audiobook?”
Sometimes the better fix is the recording itself.
What if the noise floor looks almost too good?
A very low noise-floor number is not automatically a problem.
The problem is how you got there.
Aggressive noise gates can close completely between sentences and replace natural room tone with digital silence. The meter may look impressively clean, while the listening experience becomes distracting because the background repeatedly disappears and returns.
ACX recommends maintaining room tone at the beginning and end of files and keeping overall sound and noise consistent.
For spoken word, some quiet natural room tone is normally preferable to a recording that constantly snaps between voice and absolute silence.
This is also why treating every pause and every breath as unwanted noise tends to make narration sound unnatural.
Your file sounds right, but Audio Lab still fails it
Sometimes the issue has nothing to do with processing.
ACX currently requires submitted audio to be a 192 kbps or higher constant-bit-rate MP3 at 44.1 kHz, with all files using the same mono or stereo format. Individual chapter or section files must also be under 120 minutes.
Audio Lab can flag several of these technical properties.
If the narration sounds good but the upload is failing, check your export settings before touching the audio itself.
Reprocessing a perfectly good master will not fix an incorrectly encoded MP3.
Your numbers pass, but the audiobook still has problems
This is probably the most important distinction.
ACX compliance is not just three numbers.
ACX also requires consistent sound and formatting and says files should be free from distracting sounds such as plosives, microphone pops, mouse clicks, excessive mouth noise and outtakes.
And Audio Lab does not check everything that a human quality review can reveal.
A file can therefore hit the RMS, peak and format requirements while still containing:
misreads, bad edits, obvious mouth clicks, inconsistent chapters, clipped dialogue, abrupt cuts, excessive processing or distracting background noise.
Meters are useful because they tell you something measurable.
They do not tell you whether the audiobook actually sounds good.
The order you fix things matters
A sensible troubleshooting order is:
- Listen to the raw recording first. Work out whether the problem already exists before processing.
- Remove or reduce technical problems such as noise, clicks or room issues where possible.
- Control large level differences without flattening the performance.
- Bring the chapter towards the required RMS range.
- Control peaks while preserving some dynamic movement.
- Check the finished file again by ear and against the required measurements.
This avoids the common mistake of mastering first and then discovering that every later fix changes the loudness again.
When can automated processing fix an ACX failure?
Problems such as moderate background noise, uneven levels, tonal imbalance, excessive peaks and incorrect mastering levels can often be improved during technical post-production.
That is the part of the workflow Optivox is designed to automate.
Audiobook mode runs cleanup, levelling and mastering as one repeatable technical pass and aims the finished recording towards audiobook delivery requirements rather than a generic podcast target.
But there are limits.
If the original recording is badly clipped, contains severe room echo, has major microphone problems or includes incorrect takes and misreads, mastering software cannot recreate information that was never recorded cleanly.
Likewise, technical processing should not be deciding which performance you intended to use.
Edit and proof the chapter first. Then master the finished edit.
Don't fix the meter at the expense of the audiobook
The easiest way to make audiobook mastering worse is to treat every requirement as an isolated number.
Turn the recording up until the RMS passes.
Gate everything until the noise floor looks perfect.
Compress harder because one sentence is quiet.
Limit harder because a peak escaped.
You can eventually produce a file that looks tidy on a meter and sounds noticeably processed to a listener.
The better goal is to get the technical requirements out of the way without drawing attention to the processing.
A good audiobook master should still sound like the narrator and the room they intended you to hear.
The listener should notice the story, not the noise reduction, compressor or limiter.
If you want to understand the underlying measurements in more detail, read our guide to ACX audio requirements: RMS, peaks, noise floor and file specs.
Or, if you already have an edited chapter ready to master, you can run it through Optivox audiobook mastering and compare the processed version with your original before deciding what to use.