Recording Your Own Audiobook: ACX and Audible Specs Explained

ACX — the Amazon platform that feeds Audible and Apple Books — rejects a meaningful share of first-time audiobook submissions, and almost never because the narration is bad. Files fail on measurements: RMS out of range, noise floor too high, wrong bitrate, missing room tone. The frustrating part is that ACX's requirements page states these limits in engineering shorthand and assumes you already know what they mean. This article translates every requirement into plain English and lays out a workflow that passes validation the first time.

The ACX requirements, decoded

Every uploaded file must meet all of these. Miss one and the whole submission bounces.

RMS between -23 dB and -18 dB

RMS is the average loudness of your file — not the loudest peak, the overall energy. ACX wants it between -23 dB and -18 dB so that every audiobook on Audible sits at roughly the same volume. Too quiet and listeners crank their volume between books; too hot and the narration sounds crushed. Raw home recordings usually measure around -30 dB or lower, which is why nearly every submission needs a normalization pass. A loudness-based pass with an audio normalizer raises the average level into range; the concept is unpacked further in this guide to audio normalization.

Peaks no higher than -3 dB

The single loudest instant in the file must stay at or below -3 dB. This is headroom: it guarantees no clipping after Audible's own processing. The interaction with the RMS rule is the classic beginner trap — if you normalize peaks to 0 dB, you violate the peak rule; if your delivery is very dynamic (whispers and shouts), raising RMS into range can push peaks past -3 dB. The fix is gentle compression during editing to narrow the gap between your loudest and quietest moments, then normalization.

Noise floor at or below -60 dB RMS

Record two seconds of "silence" — you sitting still, mic live. That room tone must measure -60 dB RMS or quieter. Computer fans, fridge hum, traffic, and air conditioning are the usual offenders. This is the one requirement you cannot reliably fix in software; aggressive noise reduction leaves swirly artifacts ACX also rejects. Fix it at the source: record in a small, soft-furnished space (a clothes closet is the honest industry standard), turn off the HVAC, and move the laptop outside the room or as far from the mic as the cable allows.

192 kbps or higher CBR MP3, 44.1 kHz

Deliverables must be MP3 at a constant bitrate of at least 192 kbps with a 44.1 kHz sample rate. Constant bitrate matters — VBR files get rejected even at high average bitrates. Record and edit in WAV, then convert the finished chapter with a WAV to MP3 converter at 192 kbps CBR; if a file came out VBR or at the wrong rate, a bitrate changer can re-encode it to a compliant constant 192. Background on the sample-rate side lives in this 44.1 vs 48 kHz explainer — if your interface recorded at 48 kHz, resample to 44.1 before export.

The structural rules

• Room tone padding: 0.5 to 1 second of room tone (not digital silence) at the head of each file, and 1 to 5 seconds at the tail.
• One chapter per file, each under 120 minutes, with the chapter number spoken at the top ("Chapter Seven").
• Consistent channel format: every file all mono or all stereo — no mixing. Mono is the sensible default for single-narrator work; see mono vs stereo.
• Opening and closing credits as their own files, plus a retail sample of one to five minutes.

The pass-first-time workflow

• 1. Fix the room before the gear. A $60 dynamic mic in a treated closet beats a $400 condenser in an echoey bedroom, because the condenser faithfully records the echo and the fridge.
• 2. Record WAV at 44.1 kHz. Set input gain so your loud passages peak around -12 to -6 dB. Never record "hot to be safe" — clipping at capture is unfixable.
• 3. Capture 30 seconds of room tone at the start of every session. You will use it for the head/tail padding and to patch edits.
• 4. Edit, then compress lightly. Remove mouth clicks and long gaps; apply mild compression (2:1 or 3:1) so RMS and peak targets can coexist.
• 5. Normalize to the RMS window, then verify peaks are still under -3 dB.
• 6. Export to 192 kbps CBR MP3 and spot-check the numbers before uploading. Checking one chapter end-to-end before recording twenty is the cheapest insurance in audiobook production.

Common rejection causes, ranked

• Noise floor failures — the most frequent, and the reason to solve the room first.
• RMS out of range — usually files that were peak-normalized instead of loudness-normalized.
• VBR delivery — DAW default export settings often pick VBR; always confirm CBR.
• Inconsistent files — chapter 9 in stereo when the rest is mono, or one file at 48 kHz.
• Missing room tone padding — chapters that start on the first syllable.

A note on consistency across a long project

A full-length audiobook is fifteen to thirty hours of finished audio recorded across weeks, and ACX evaluates the files as a set. The subtle failure mode is drift: you move the mic, the seasons change the room, you buy a new interface halfway through — and suddenly chapter 19 sounds different from chapter 3 even though both pass the meters individually. Guard against it mechanically. Mark the mic position with tape, save your recording and processing settings as a preset, and re-record a short reference sentence at the start of each session to compare by ear against the first chapter. If a retake session months later sounds noticeably different, it is almost always the room or the distance, not the gear — and it is far cheaper to match the setup than to re-record the chapter.

None of these require a mastering engineer. They require knowing the numbers, a quiet room, and a consistent export chain — record WAV, edit, normalize, convert to 192 kbps CBR — applied identically to every chapter. Do that and validation becomes a formality instead of a lottery.