Loudness outside spec
Files measured even slightly outside -23 to -18 dBFS RMS or the -3 dBTP peak ceiling can be rejected in technical review.
Publishing an audiobook means five decisions, in order: confirm you own the rights, get your manuscript ready, choose a narrator or an AI voice, decide how widely to distribute, and avoid the technical mistakes that get a first submission rejected. Here is each step, in order.
September 19, 2026 · 10 min read
For a self-published title, audio rights are normally negotiated separately from print and ebook rights - most self-published authors already hold them by default. If you signed a traditional publishing contract at any point, check it before producing anything: some contracts carve out audio rights specifically. You need to know you own the right to sell an audiobook version before you spend on narration.
An EPUB or DOCX file with chapters clearly marked is enough to start, whether you are hiring a narrator or producing with AI. Clean up anything that reads oddly aloud - unusual abbreviations, invented names, numbers written as digits - before recording begins, human or AI.
This decision determines which stores you can reach. A human narrator, hired through ACX, keeps ACX and Audible open to you. An AI voice is faster and does not require an audition, but ACX requires a human narrator; the only synthetic voices it takes are Amazon's own (Virtual Voice, through KDP). Other stores set their own rules, and some accept AI narration once you disclose it - that is the route an AI-produced book takes instead.
ACX expects MP3 at 192 kbps, RMS between -23 and -18 dB, peaks at -3 dB, and 1 to 5 seconds of room tone (its recommended range) at the head and tail of each chapter file; other stores publish their own loudness targets. Confirm your files meet this before submitting anywhere - it is a common reason a first submission gets bounced back.
If you went with a human narrator, ACX gives you a choice: exclusive (50% royalty under ACX's current model, Audible, Amazon and Apple Books only) or non-exclusive (30% royalty, sell on Spotify, Kobo and your own site at the same time). Under ACX's legacy agreement, exclusivity ran seven years from your first sale; check the current term in your contract. An AI-narrated book, going to the stores that accept AI narration, is not an ACX submission at all - there is no exclusivity to opt into.
Human narrator. Costs $200 to $500 per finished hour (roughly $2,000-$5,000 for a 10-hour book) or a royalty share instead of an upfront fee. This is the only path onto Audible through ACX, since ACX requires a human narrator.
AI voice. No per-hour rate and no audition - you pick a voice, approve a sample from your own pages, and the book is produced in about an hour. Audible through ACX is not available on this path. The files go to the stores that accept AI narration instead, with the disclosure they require handled for you.
Files measured even slightly outside -23 to -18 dBFS RMS or the -3 dBTP peak ceiling can be rejected in technical review.
Stores that accept AI narration require a written disclosure. Skipping it, or assuming it is optional, can get a listing pulled after it is already live.
ACX requires a human narrator; the only synthetic voices it takes are Amazon's own (Virtual Voice, through KDP). Producing an AI-narrated book and then trying to submit it to ACX wastes the production run entirely.
Whichever path you choose, hear it before you commit to a full production run. Get a chapter of your own manuscript narrated free, delivered within 48 hours, no commitment attached.
These five steps look the same whether you are publishing one book or ten. The cost structure does not. A human narrator is priced per book - the $2,000-5,000 range repeats in full for every title, on its own timeline of weeks to months. An AI production plan is priced per month of output instead: once you know your page count for the year, the cost per additional title drops with every book you produce inside that plan.
That difference matters most for authors with a backlist or a series in progress. Publishing a single standalone title rarely changes which path makes sense. Publishing three or four books a year usually does - the fixed per-book cost of a human narrator adds up in a way a flat monthly plan does not.
The exclusivity decision compounds the same way. Choosing ACX exclusive on a first book locks that title to Audible for the length of the agreement - seven years under the legacy contract, so confirm the current term - a decision made once, felt for years. Choosing an AI voice for a backlist sidesteps that question for every title in it, since there is no exclusivity term to weigh in the first place. Neither choice is wrong on its own; the point is to make it once, deliberately, rather than book by book without noticing the pattern.
Confirm you hold the audio rights, get your manuscript ready (EPUB or DOCX, chapters clearly marked), choose a human narrator or an AI voice, master the files to each platform's spec, then submit yourself through ACX, Apple Books for Authors, or an aggregator like Findaway Voices. No publisher or agent is required for any of these steps.
Usually, yes, for a self-published title - audio rights are normally negotiated separately from print and ebook rights, so most self-published authors already hold them. If you signed a traditional publishing contract at any point, check it: some contracts carve out audio rights specifically, and you need to confirm you were not one of them before you produce or sell an audiobook.
Not through ACX. Audible's ACX requires a human narrator for the audiobooks you upload yourself; the only synthetic voices it takes are Amazon's own (Virtual Voice, through KDP). Other stores set their own rules, and some accept AI narration once you disclose it - that route does not go through ACX or Audible.
Files outside ACX's loudness window. ACX expects MP3 at 192 kbps, RMS between -23 and -18 dB, peaks at -3 dB, and 1 to 5 seconds of room tone (its recommended range) at the head and tail of each chapter; other stores publish their own loudness targets. A file that falls outside that range can be rejected in technical review.
Choose your path, hear your book narrated, then publish on your own terms.