How to Create an ACX Audiobook From Your Ebook in 2026: The Complete Author Guide

What ACX actually requires (the full spec)

Every audiobook submitted to Audible through ACX has to clear the same automated QC bar before a human ever listens to it. Here is the list from ACX's audio submission requirements (updated 15 April 2026), with what each line item means in practice:

The specs that most often trip first-time uploads are sample rate, bitrate type (CBR vs VBR), mixed channel formats, the RMS loudness window, and per-chapter file structure. A file that is 48 kHz / 192 kbps VBR / -16 RMS fails three of them in the same upload.

Why hitting every spec by hand is hard

The list above looks tractable until you sit down to produce the master. The two failure modes that catch most authors:

Loudness normalization is not the same as gain. ACX wants the long-term average loudness of every file to land between -23 and -18 dBFS RMS. That is not a setting in most consumer audio editors. You need a loudness normalization tool that targets integrated LUFS or RMS specifically, and you need to apply it after recording, not during. Authors who simply boost gain or normalize peaks end up with files that pass the peak check but fail the RMS check.

The CBR vs VBR distinction is invisible in most software. When you hit "Export to MP3" in a default editor, the output is usually VBR (variable bitrate) because it produces smaller files at the same perceived quality. ACX rejects VBR submissions on principle: their player has historically had seek bugs on VBR files. You have to specifically configure the encoder to output 192 kbps CBR with min and max bitrate locked, which is buried in advanced export settings and not always exposed in consumer tools.

Add per-chapter file splitting (every chapter as its own MP3, named in track order, each under 120 minutes), retail sample creation, opening/closing credits, and noise floor checks, and a single audiobook submission can take an experienced author 6 to 10 hours of post-production after the recording is done. Inexperienced authors often spend longer and still miss the spec, learning each rule one rejection email at a time.

Two paths to ACX-spec audio

There are essentially two production paths in 2026:

Path A is the right choice when narration performance matters more than economics: dialogue-heavy fiction with multiple characters, dramatic readings, comedy timing. Path B is the right choice when consistency, speed, and cost matter more than performance: non-fiction reads where a clear, neutral, professionally paced voice is what the audience expects, sold on the stores that accept AI narration. If the book must be on Audible through ACX, only Path A qualifies. Our genre-by-genre AI vs human narrator comparison covers which fits which book in detail.

The streamlined path: manuscript to ACX-spec zip

For an author shipping a non-fiction or memoir audiobook with the AI-native path, the actual workflow inside Inkfluence AI is six steps:

1. Write or import the manuscript. If you already have a manuscript, the document importer accepts DOCX, PDF, and EPUB and detects chapters automatically. If you are starting from scratch, the AI book generator drafts the full manuscript first.
2. Open the audiobook studio. Each project has an Audiobook tab that lists every chapter with a "Generate" button. Pick a voice from the catalogue, set the speaking rate (180 words per minute is the audiobook standard), and click generate.
3. Review and re-record. Generated chapters play back inline. If a chapter has a mispronunciation or a section reads awkwardly, edit the manuscript text and re-record that chapter only. Other chapters are not affected.
4. Prepare the chapter ZIP. This is the difference between the single merged book MP3 and a set of chapter files at ACX's technical spec. The chapter export runs every chapter through the loudness normalization chain (RMS about -20 dB, peaks below -3 dB), resamples to 44.1 kHz, encodes as 192 kbps CBR mono MP3, adds room tone at the head and tail plus opening and closing credits files, and bundles the result as a zip with chapter-numbered file names.
5. Download the zip and publish. Each MP3 inside is named in reading order ("01 - Chapter Title.mp3", "02 - Chapter Title.mp3", and so on). Upload them to a store that accepts AI narration, such as Google Play Books or Kobo, and mark the narrator as a synthesized voice where the store asks. They are not for ACX, which requires a human narrator unless it has authorised otherwise. Where to sell an AI audiobook lists the stores that take them.
6. Prepare a retail sample if the store asks for one. A few minutes from the opening of chapter 1 is the usual choice; cut it from the chapter file.

The full pipeline runs in about 10 to 30 minutes for an 8-chapter, 4-hour audiobook. The longest single step is voice selection and chapter generation, which is unattended (you start it, walk away, come back). Mastering, normalization, and packaging that would take an audio engineer 4-6 hours happens in seconds because it is all scripted in the export pipeline.

The seven most common ACX rejection reasons

From talking with self-published authors who have submitted to ACX since 2022, the same handful of issues come up over and over. In rough order of frequency:

1. RMS loudness outside the -23 to -18 dBFS window. The single most common automated-QC failure. The fix is a loudness normalization pass targeting -20 dBFS RMS as the safe centre of the window.
2. Sample rate is 48 kHz instead of 44.1 kHz. Default for many recording interfaces. The fix is resampling the master before export.
3. Mixed channel formats. Some chapters mono, some stereo. ACX wants every file in the same format; the usual fix is downmixing every chapter to mono with equal weighting from both channels.
4. VBR encoding instead of CBR. Default for most "Export to MP3" buttons. The fix is configuring the encoder explicitly for 192 kbps constant bitrate.
5. Background noise above -60 dBFS. Caused by an untreated recording space, fan noise, or a too-hot mic gain. The fix is room treatment plus a noise reduction pass.
6. Chapters merged into a single file. The author exported the entire audiobook as one MP3. The fix is splitting on chapter boundaries before export.
7. Opening or closing credits missing. ACX expects opening credits that state the title, author and narrator, and closing credits that make the ending clear, such as "You have been listening to Title, by Author, narrated by Narrator. The End."

Path B sidesteps the technical ones by running every export through the same scripted chain, but it does not get an AI-narrated book past ACX's narration rule, which is not a technical check. Path A authors typically learn each rule one rejection email at a time, which is why "studio mastering for ACX" is itself a $50-to-$200 service offered by audio engineers on Reedsy and elsewhere.

Cost: studio versus AI-native production

Three real-world scenarios for an 8-hour non-fiction audiobook (about 80,000 words narrated at 180 wpm), with all costs as of 2026:

For the breakdown of AI vs human narration economics (and which genres each fits), see our 2026 audiobook narrator comparison. For the full self-publishing budget context (where audiobook fits next to editing, cover, and ISBN), see the real cost of publishing a book in 2026.

Royalty share, per-finished-hour, or DIY: which makes sense

ACX gives authors three commercial structures when working with a human narrator:

• Per-Finished-Hour (PFH). Author pays the narrator a flat rate per hour of finished audiobook, typically $200 to $600. Author keeps 100% of royalties. Best for authors with cash on hand and confidence in book sales.
• Royalty Share (RS). No upfront payment. Narrator and author split net royalties 50/50 forever. Best for first-time authors or low-cash budgets, but the royalty split lasts the life of the book on Audible.
• Royalty Share Plus. A reduced upfront PFH rate (often $100 to $200/hr) plus a royalty share. Splits the risk between author and narrator.

Authors who AI-narrate pay no narrator and keep whatever the store pays, but they sell outside ACX, on the stores that accept AI narration. The economics tilt toward AI narration when a hired narrator's fee is likely to exceed what the audiobook will earn. For genres where a professional narrator measurably moves sales (literary fiction, character-heavy thrillers, emotional memoir), PFH or RS Plus are still the right call.

Frequently asked questions

Related resources