ACX requires human narration by default. Its submission requirements state plainly that "your submitted audiobook must be narrated by a human unless otherwise authorized," and that "unauthorized use of text-to-speech, AI, or automated recordings in ACX titles is prohibited."[1] So the short answer for an independent author or narrator in June 2026: you cannot drop a generic AI voice into an ACX title and submit it. A human reads the book, with one narrow, narrator-controlled exception covered below.
This is the policy question, not the technical one. The ACX audio requirements tell you the loudness, peak, and noise floor numbers a file must hit. They do not tell you who, or what, is allowed to narrate. This page covers eligibility: the rule on the voice itself, what changed in 2025 and 2026, and what is still genuinely unsettled.
Does ACX allow AI narration in 2026?
Not for the standard independent route. The default rule is human narration, and unauthorised AI or text-to-speech is prohibited.[1] The word doing the work is "unauthorized." ACX has opened one authorised path, and it is not a free pass to generic synthetic voices.
That path is the Narrator Voice Replica beta, announced on 9 July 2025.[2] It lets a small, invited group of professional narrators create a computer-generated replica of their own voice from their recordings, then use it on projects they choose. The control sits with the narrator throughout:
- They opt in by request; the beta invites "a small group of narrators to participate."[2]
- "Audible will not separately use a narrator's voice replica for any content without their approval."[2]
- They "always choose the projects they want to audition for and accept," title by title.[2]
- Titles narrated with a replica are "labelled in the narrator field of the title's listing and detail page," so listeners can see it.[2]
So the authorised AI on ACX is a narrator's own voice, used with their consent, and disclosed. It is not a way to skip hiring a narrator. It launched as a US-only beta,[2] and as of spring 2026 it remains in beta, accessible to participating narrators rather than open to every author.[3]
The distinction worth holding onto: the replica is a tool a human narrator extends their own work with, not a substitute for one. A narrator who joins the beta still reviews and edits every replica-produced title for pronunciation and pacing before it ships.[2] If you are an author hoping to type a manuscript into a synthetic voice and list it on Audible through ACX, this beta is not that. The narration still originates from a real person who has agreed to it.
What about Audible's AI narration programme?
This is where most of the confusion comes from, because two different things share the word "Audible."
On 13 May 2025, Audible announced it would expand its catalogue using AI narration and AI translation, offering publishers "more than 100 AI-generated voices" across English, Spanish, French, and Italian.[4] That announcement is aimed at publishing partners, not at the independent authors and narrators who use ACX. The newsroom post does not mention ACX at all.[4]
It is worth holding the two apart:
| Channel | Who it serves | AI narration in 2026 |
|---|---|---|
| ACX | Independent authors and narrators | Human by default; narrator voice replica beta is the only authorised AI route[1][2] |
| Audible publisher programme | Publishing partners working directly with Audible | 100+ AI voices offered through Audible-managed or self-service production[4] |
If you publish through ACX, the publisher AI voices are not your route, and the human-narration rule still governs your title. The two programmes run on separate tracks.
What is confirmed, and what is still in flux?
Accuracy matters more than a tidy answer here, because the policy has moved twice in roughly a year. Here is the honest split as of June 2026.
Confirmed. ACX requires human narration unless otherwise authorised, and unauthorised TTS or AI is prohibited.[1] The narrator voice replica beta exists, is opt-in, is narrator-controlled, and labels affected titles.[2] Audible's separate 100-voice AI programme is for publishers, not ACX authors.[4]
In flux. ACX's own requirements note that it is "working to accept third-party TTS content from publishers and creators," with no published timing.[1] The voice replica programme is still a beta, so its terms and availability can change.[3] If you are planning a project months out, check the ACX submission requirements again before you commit, because the line on AI is the part of ACX policy most likely to shift.
The direction of travel is clear enough: in barely a year, ACX went from a flat human-only rule to running an authorised, narrator-controlled AI route, while its parent Audible built a separate publisher programme with more than a hundred AI voices.[2][4] What is not clear is where the independent author lands when that "working to accept third-party TTS" line eventually resolves, or on what terms. Treat any blog post or forum thread claiming ACX now freely accepts AI narration with caution, because the only authorised independent route in June 2026 is the voice replica beta, and that route still starts with a consenting human narrator.[2]
We will not pretend this is settled. What we can say with confidence is the part that does not move: the technical bar.
Why the technical spec applies whatever the policy decides
Here is the practical point for anyone producing a file. Whoever or whatever narrates, the audio still has to pass the same automated check.
ACX measures every submitted file against a fixed set of technical specs before a human ever listens: loudness, true peak, noise floor, sample rate, channel count, format, and the head and tail silence.[1] Those numbers do not change based on who held the microphone. A human-recorded chapter, a labelled voice replica, and a publisher's AI-narrated title are all judged against the same loudness window and the same noise floor ceiling. The full ACX audio requirements list every one.
This is why the policy question and the production question are separate problems. The policy decides whether a voice is eligible. The spec decides whether the recording of that voice is submittable. A perfectly legitimate human reading still gets returned if its RMS sits below the floor or its peaks cross -3 dBFS. Eligibility does not exempt anyone from the measurements.
So the work of getting a file to spec is the same regardless of where the policy lands. Loudness has to sit in range. Peaks have to stay under the ceiling. The noise floor has to stay quiet enough that the gain you add to reach loudness does not drag it over the line. That last tension, the gain-budget trap, catches recordings from every source and is covered in the audiobook noise floor guide.
Where does chapterpass sit in all of this?
chapterpass does not narrate, and it does not generate voices. It is a mastering tool. It takes a finished recording, human-read or otherwise, and masters it to ACX's technical spec: loudness, peak, noise floor, format, and silence padding, then verifies every value against its target.
That makes it orthogonal to the narration policy. chapterpass does not decide, and does not care, whether your voice is eligible under ACX's rules; that is your call and ACX's. What it does is take whatever you have already recorded and bring the numbers into range so the file passes the automated check. If ACX widens its AI policy tomorrow, the mastering job is unchanged, because the spec is unchanged.
Want to see where your recording stands against every ACX measurement before you submit? Run the Free ACX Check. The file stays in your browser, there is no account, and the report shows each value against its ACX target.