ACX Audio Requirements Explained: RMS, Peak and Noise Floor

Every audiobook submitted to ACX is checked by a human quality-assurance reviewer against a fixed set of technical numbers. Those numbers are not guidelines. A file that misses one of them comes back, however good the narration is.

This is the full list, what each number actually measures, and which ones cause the most rejections.

The ACX technical requirements in full

What each number means

RMS: between −23 dB and −18 dB

RMS is average loudness measured across the whole file. It is the number that decides whether your audiobook sits at a comfortable volume next to every other audiobook in a listener’s library.

The window is five decibels wide, which sounds generous until you record across several sessions. A narrator who is tired on Tuesday and energetic on Thursday can drift outside it without hearing the difference. RMS is measured per file, so a book that passes on chapter one can still fail on chapter nine.

The fix is not to record more consistently — nobody does. It is to measure every file and correct the ones that drift.

Peak: no higher than −3 dB

Peak is the single loudest instant in the file. The −3 dB ceiling exists to leave headroom for the MP3 encoder, which can push a sample slightly above where it sat in the uncompressed file.

Peaks are usually caused by plosives, a laugh, a cough, or a chair. One stray sound three hours into a recording is enough to fail the whole file.

Noise floor: −60 dB RMS or quieter

This is the one that fails most books.

The noise floor is whatever is present when the narrator is not speaking: air conditioning, a computer fan, traffic, the hiss of the preamp itself. −60 dB is quiet. It is quieter than most spare rooms, and considerably quieter than most USB microphones running at high gain.

Two things make this harder than it looks. First, raising RMS to reach −18 dB raises the noise floor by exactly the same amount, so fixing one number can break the other. Second, aggressive noise removal does pass the measurement, but it leaves the voice sounding thin, phasey or underwater — and ACX rejects for that too, under audio quality rather than noise floor.

The honest position: a noise floor problem is a recording problem. It can be reduced in post, but only so far before the voice starts to suffer.

Room tone at head and tail

Room tone is the silence of your actual recording space, not digital silence. ACX asks for 0.5 to 1 second before the first word and 1 to 5 seconds after the last.

Digital silence — a flat line at zero — is a fail. The transition from true silence into room noise is audible as a small click or swell, and reviewers listen for it.

The rejections that happen most

1. Noise floor above −60 dB. Almost always a room or microphone problem rather than an editing one.
2. Over-processing. Noise reduction pushed far enough to meet the measurement, at the cost of a natural voice.
3. Inconsistency between files. Chapter three recorded closer to the microphone than chapter four.
4. Digital silence instead of room tone. Usually the result of an editor trimming with a noise gate.
5. Wrong file format. Variable bitrate MP3, 48 kHz, or a single file for the whole book.
6. Missing opening and closing credits. These are separate files, not part of chapter one.
7. Mouth noise and breaths left in. Not a measured specification, but reviewed subjectively and a common reason for a return.

How to check your files before submitting

Audacity’s free ACX Check plug-in measures all three of the critical numbers and tells you plainly which fail. Run it on every file, not a sample. It takes a few seconds per chapter and catches almost everything a reviewer would.

If a file fails, note which number failed before changing anything. Raising a quiet file to fix RMS will raise the noise floor with it, and the two have to be solved in the right order: clean first, then set level, then check peaks.

What happens after a rejection

ACX returns the book with a note naming the problem. You correct and resubmit. There is no penalty for a rejection beyond the time it costs, but review queues mean each round trip can add one to two weeks to a release date.

That is the real cost of a failed submission: not the fix, but the calendar.

Getting it right the first time

Most ACX failures come down to measurement rather than skill. The narration is fine; nobody checked the numbers until after the book was finished.

If you would rather hand that part over, our audiobook editing service masters to the full ACX specification and proof listens against the manuscript before anything is submitted — so the first file you send is the one that passes.