ACX is the Audible platform where authors, narrators and producers make and submit audiobooks. Its audio requirements cover levels, noise, encoding and how the book is split into files, and one missed detail can send a title back for corrections. This guide goes through every current rule, explains what it means and how to meet it, and ends with a checklist to run before you upload.
The rules are taken from ACX's current requirements page (April 2026) and glossary. Where ACX is silent, we say so instead of guessing.
The three level requirements
ACX sets three level limits for every file: RMS, peak and noise floor. All are plain decibel readings relative to digital full scale; none is a LUFS figure, and ACX publishes no LUFS target for audiobooks.
RMS between -23 dB and -18 dB
Each file must fall between -23 dB and -18 dB RMS. RMS (root mean square) is the average power of the file. The ACX Check plugin for Audacity measures it unweighted and over the whole file, pauses included, so a narration with long silences reads lower than the same narration edited tightly.
Aim near the middle, around -20 dB, so small differences between meters cannot push you over an edge. ACX itself warns that readings "may differ depending on your DAW, the plugin/meter used". One cause is calibration: ACX Check reads a full-scale sine wave as about -3 dB RMS, while meters following the AES17 convention show 0 dB, so one file can read about 3 dB apart in two tools.
There is no exact conversion between this RMS window and LUFS. LUFS vs RMS vs peak explains why.
Peaks no higher than -3 dB
Each file must have peak values no higher than -3 dB. ACX's glossary defines peak as "the maximum instantaneous level of a signal" but never says whether that means the highest sample (sample peak) or the highest point of the reconstructed waveform (true peak). The ACX Check plugin measures sample peak.
The safe approach covers both readings: keep the true peak at or below -3 dBTP. True peak is never lower than sample peak, so that file also passes a sample-peak check. Leave a little extra margin, because MP3 encoding can raise peaks, and measure the MP3 you upload, not the WAV you exported. True peak vs sample peak covers the details.
Noise floor no higher than -60 dB RMS
Each file must have a noise floor no higher than -60 dB RMS. ACX does not publish how it measures the floor; it is judged during ACX's human quality review. The ACX Check plugin measures the RMS of a 400 ms window stepped every 100 ms across the selection and reports the quietest value, which its documentation describes as the quietest 500 milliseconds.
The floor rises with every decibel of gain you add to reach the RMS window, so measure it on the finished file, not the raw recording. Getting an audiobook noise floor below -60 dB covers measurement and fixes.
Numbers are only part of the review. ACX's Audio Lab checks RMS, peak levels, bit rate, bit-rate method, sample rate, mixed channels, duplicate files and spacing. It does not detect noise floor or editing errors; people listen for those.
File format and encoding
MP3, 192 kbps or higher, constant bit rate, 44.1 kHz
Every file must be "a 192 kbps or higher CBR, 44.1kHz MP3." ACX requires constant bit rate (CBR) so its encoding process works without error, which rules out variable and average bit-rate files. You may upload 256 or 320 kbps, but ACX notes that the difference heard by listeners will be negligible. A project recorded at 48 kHz has to be resampled to 44.1 kHz before encoding.
No joint stereo
The ACX glossary states that "ACX does not accept Joint Stereo files." Joint stereo is an MP3 mode that encodes the two channels together to save bits, and some encoders switch it on by default for stereo; ffmpeg's LAME-based MP3 encoder is one of them. If you deliver stereo, set the encoder to plain stereo. Mono files are unaffected. The file formats guide shows the exact setting.
All files in the same channel format
All files in a title must use the same channel format: all mono or all stereo. The current requirements page does not prefer either. The well-known line that ACX "strongly recommends mono" comes from a 2014 ACX blog post and is not part of the current requirements. Mono vs stereo explains the trade-offs.
The current page states no bit depth and no file-size limit.
How the book is split into files
One chapter or section per file, at most 120 minutes
Each file must contain only one chapter or section, uploaded individually, and no file may be longer than 120 minutes. If a chapter runs longer, split it into separate files and include a secondary header for continuity, for example by announcing the part number at the start of the second file.
Opening and closing credits
Opening and closing credits must be separate files. The opening credits must state the audiobook's title, author or authors, and narrator or narrators. The closing credits must indicate finality, so the listener knows the book has ended. Music is allowed in the credits as long as it does not obscure the narration.
Retail sample
You also provide a retail sample of five minutes or less. ACX asks you to use the first five minutes of the audiobook, and the sample must not contain explicit material.
Room tone and spacing
ACX recommends "between 1 and 5 seconds of room tone at the beginning and end of each file" and states that "room tone spacing must not exceed 5 seconds", so no pause should run past five seconds. Room tone means the natural sound of your recording space, not digital silence. The older advice of half a second to one second at the head comes from a 2017 ACX blog post and has been replaced.
Narration and consistency
Your audiobook must be narrated by a human unless ACX has authorized otherwise. ACX prohibits unauthorized use of text-to-speech, AI or automated recordings.
The book must also be consistent in sound and formatting: audio levels, tone, noise level, spacing and pronunciation should match from file to file, and ACX asks for files free of plosives, mic pops, clicks and outtakes. A chapter recorded on another day with the microphone moved can pass every measurement and still sound out of place.
Typical reasons files fail
- RMS too low. Raw narration often sits below -23 dB RMS, and plain gain pushes peaks past -3 dB, so the level has to come up with compression or limiting.
- Peaks over the limit after encoding. A WAV limited to exactly -3.0 dB can measure slightly higher once it is an MP3.
- A floor pushed up by levelling. Narration at -26 dB RMS with a -64 dB floor ends up with a -58 dB floor once it is raised 6 dB to -20 dB RMS.
- Export settings. Joint stereo, variable bit rate, or a 48 kHz project exported without resampling.
- Mixed channel formats. A few stereo chapters in an otherwise mono book.
- Digital silence. Pauses cut to absolute silence sound unnatural, and the ACX Check plugin warns when the floor falls below -90 dB.
- Structure. Two chapters in one file, credits attached to the first chapter, or a file over 120 minutes.
ACX submission checklist
| Requirement | Rule | How to check |
|---|---|---|
| RMS level | -23 to -18 dB per file | Unweighted RMS, whole file |
| Peak | No higher than -3 dB (aim for a true peak at or below -3 dBTP) | Measure the encoded MP3 |
| Noise floor | No higher than -60 dB RMS | Quietest 400 ms of the finished file |
| Encoding | MP3, CBR, 192 kbps or higher, 44.1 kHz, no joint stereo | Encoder settings, media info |
| Channels | All files mono, or all stereo | File properties |
| Length and sections | One chapter or section per file, 120 minutes at most | File list and durations |
| Room tone | 1 to 5 seconds at head and tail; no gap over 5 seconds | Listen, inspect the waveform |
| Credits | Separate opening and closing credit files | File list |
| Retail sample | 5 minutes or less, from the start, nothing explicit | Listen |
| Narration | Human, unless authorized by ACX | Production records |
The free ACX check measures RMS, peak and noise floor in your browser, using the same 400 ms window as the ACX Check plugin, so you can test a chapter before and after mastering.
A simple mastering chain for ACX
Audacity's own audiobook mastering guide uses a short chain: a low roll-off filter for speech, RMS normalization to -20 dB, a soft limiter at -3.5 dB, and then the ACX Check plugin. It adds that if the noise is quieter than about -65 dB, you may be done. The same order works in any editor: remove rumble, reduce steady noise, bring the narration into the RMS window, limit the peaks with some margin, encode to 192 kbps CBR MP3 at 44.1 kHz, then check the MP3.
If you would rather not assemble that chain yourself, LufsReady's ACX audiobook preset aims for -20 dB RMS, limits true peak to -3.5 dBTP, measures the noise floor the way ACX Check does and encodes 192 kbps CBR MP3 at 44.1 kHz without joint stereo, with a pass or fail per check in its report. Meeting the numbers is necessary but not sufficient: ACX's reviewers also listen for consistency, editing and noise.
Sources
- ACX: Audio Submission Requirements
- ACX: Audio terminology glossary
- ACX: Audio Lab (audio analysis tool) FAQ
- Audacity: Analysis plugins (ACX Check)
- Audacity: Audiobook mastering guide
Check your file in seconds
Measure loudness, true peak and noise floor for free in your browser, or master a file to the exact target with a report.