ACX audio requirements explained, and how to fix a failed file

Most audiobooks that come back from ACX review do not fail on the performance. They fail on a handful of numbers: the file is too quiet, the peaks are too hot, the room is too noisy, or the silence at the start is the wrong length. Every one of those can be measured before you upload, and every one has a standard fix.

This guide explains each requirement in plain words, why files miss it, and what to change. If you want to see where your own files stand first, drop them into the free audiobook file checker. It runs in your browser and nothing is uploaded.

The requirements at a glance

Requirement	ACX value	What it protects
RMS loudness	-23 to -18 dB	Every book plays at a similar volume
Peak level	-3 dB or lower	No distortion on loud words
Noise floor	-60 dB or lower	No hiss or hum in the pauses
Room tone	1 to 5 s at the start and end	Clean gaps between chapters
File format	MP3, 192 kbps or higher, constant bit rate, 44.1 kHz	Files that every player handles
Length	Up to 120 minutes per file, one chapter per file	Easy navigation for listeners
Channels	All mono or all stereo	No jumps between chapters

Spotify for Authors and most other distributors ask for the same format and very similar levels, so a file that meets these numbers is ready for most stores.

RMS: the average loudness

RMS is the average level across the whole file, pauses included. ACX wants it between -23 and -18 dB. Below that range listeners have to turn the volume up, and above it the narration starts to sound squashed and tiring.

Why files fail: most home recordings come out quiet, often around -26 to -30 dB, because narrators leave headroom when they record. That is a good habit, it just means the level has to be brought up afterwards.

The fix: raise the overall level with gain or a loudness normalizer until RMS sits near -20 dB, the middle of the window. A light compressor before the limiter makes the level more even and helps a quiet file reach the window without pushing the peaks. Avoid peak normalizing to -3 dB as the main tool: it only sets the loudest moment and often leaves the average too low.

Peak level: the loudest moment

No single sample may go above -3 dB. Loud words, laughter and hard consonants are where it happens.

The fix: put a limiter last in your chain with its ceiling at -3.5 dB. The extra half decibel matters, because converting to MP3 can push a peak slightly higher than it was in the WAV. If your RMS is already near the top of its window, lower the gain a little instead of limiting harder.

Noise floor: the sound of your room

The noise floor is the level of the recording when nobody is speaking: computer fans, traffic, the hum of a fridge, the hiss of the microphone preamp. ACX wants it at -60 dB or lower. Checkers usually measure the quietest half second of the file.

Why files fail: this one is tied to RMS. If you recorded quietly and then raised the gain by 8 dB to fix RMS, the noise came up by 8 dB too. A room that measured -66 dB in the raw recording ends up at -58 dB and fails.

The fix: first reduce the cause, since it is far easier than fixing it later. Record closer to the microphone, turn off what you can, and add soft furnishings. Then, if needed, capture a few seconds of room tone, use it as the profile for gentle noise reduction, and keep the reduction light so the voice does not sound watery. A noise gate or downward expander between sentences also helps, as long as it does not cut to pure digital silence.

Room tone at the start and end

Each file needs between 1 and 5 seconds of room tone before the first word and after the last. It gives a clean pause between chapters and between the credits and the book.

The fix: keep a recording of your room tone and paste it in where you need it. Do not use generated digital silence: a stretch of absolute zero sounds like a dropout, especially on headphones. If your editor trimmed too tightly, add a little room tone back; if a chapter starts with ten seconds of breathing and page turns, trim it.

Format, length and consistency

• MP3 at 192 kbps or higher, constant bit rate. Variable bit rate files are rejected even when the average is high. Look for CBR in your export settings.
• 44.1 kHz. If you recorded at 48 kHz, convert when you export the final MP3.
• One chapter per file, up to 120 minutes. Split a very long chapter at a natural break and give each part its own room tone.
• Same channel format for every file. Mono is the usual choice for narration. Do not mix mono and stereo chapters.
• Keep a WAV master. Fix problems on the WAV and export a new MP3, rather than re-encoding an MP3 that already lost detail.

A mastering order that works

1. Edit the performance: remove mistakes, long breaths and clicks.
2. Apply light noise reduction if the room needs it.
3. Use a gentle EQ to remove low rumble below about 80 Hz.
4. Raise the level to about -20 dB RMS, with a light compressor if the level jumps around.
5. Put a limiter last, with its ceiling at -3.5 dB.
6. Check room tone at the start and end, then export MP3 at 192 kbps CBR, 44.1 kHz.
7. Run every chapter through the checker, then listen on headphones.

The numbers are what a machine can check. ACX also listens for mouth noise, plosives, background sounds, edits you can hear and chapters that do not match. A file that passes every check still deserves a full listen.

What about AI narrated audiobooks?

ACX requires narration by a human, apart from the virtual voice program Amazon runs itself, so AI narration made with other tools is not accepted there. Spotify for Authors, Google Play Books and Kobo do accept digital narration when you disclose it. VoicedBook narrates your manuscript and masters every chapter to the levels above automatically: about -20 dB RMS, peaks at -3 dB or lower, a room tone bed well under -60 dB, 1 second of room tone at the start and 3 at the end, and MP3 at 192 kbps CBR, 44.1 kHz. Each export is checked against these numbers before you download it.

Common questions

What are the ACX audio requirements?

MP3 at 192 kbps or higher with a constant bit rate and a 44.1 kHz sample rate, RMS between -23 and -18 dB, peaks at -3 dB or lower, a noise floor of -60 dB or lower, 1 to 5 seconds of room tone at the start and end of each file, one chapter per file and up to 120 minutes per file.

What is the most common reason files fail?

Low RMS together with a high noise floor. A quiet recording has to be raised to reach -23 dB, and the gain raises the room noise with it. Record closer to the microphone and in a quieter room, so the fix is small.

Should I normalize to -3 dB peak?

No. Peak normalizing sets the loudest moment and leaves the average where it was, so the file often stays too quiet. Set the average level first, around -20 dB RMS, and then use a limiter at -3.5 dB to catch the peaks.

Is digital silence allowed between sentences?

The requirement is about room tone, not silence. Pure digital silence between sentences sounds like the audio dropped out, and reviewers flag it. Use a gentle noise gate or expander instead of hard muting.

Can I check my files before uploading?

Yes. The free VoicedBook audiobook checker measures every one of these numbers in your browser, without uploading the files, and explains how to fix each failure.