Audio & Voice
ACX bans AI narration: where a voice clone still pays
You cannot publish an ElevenLabs-narrated audiobook on ACX. Its submission requirements state 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.” Two authorized AI routes do exist — and the structural catch is that both of them generate the voice themselves, so neither will take a clone you bring. This page replaces an earlier version of ours that told you to do exactly the thing ACX prohibits.
What This Covers
The short version What ACX actually requires Both AI routes supply the voice The royalty model changed, and it expires Why 50% can pay less than 40% The seven-year lock is not seven years Where a voice clone does work What we could not verify FAQThe short version
Sources: the ACX audio submission requirements, how royalties work and new royalty model pages, Amazon audiobook royalties, Spotify on digital voice narration and the ElevenLabs PVC documentation, all read September 2026.
What ACX actually requires
ACX publishes a list of audio submission requirements — bit rate, peak level, noise floor — and at the end of the technical specifications sits a sentence that decides whether any of the rest matters to you:
“Your submitted audiobook must be narrated by a human unless otherwise authorized. Unauthorized use of text-to-speech, AI, or automated recordings in ACX titles is prohibited.”
Read the hinge word. The rule is not whose voice it is, it is whether the use is authorized. Much published advice assumes cloning your own voice sidesteps the problem, on the reasoning that it is still, in some sense, you narrating. Nothing in that sentence supports the distinction: a synthetic rendering of your own voice is a text-to-speech recording, and unless Amazon authorized the route you used, it falls inside the prohibition.
This site published the opposite advice. An earlier version of this page walked readers through room treatment, microphone selection, Professional Voice Cloning and then submission to ACX as one workflow, and it was wrong at the last step. We retract that here rather than quietly deleting it, because the old headline is what anyone arriving from search clicked on.
ACX also pushes the duty to check downhill: it tells rights holders that while it monitors narrators for text-to-speech work, it is ultimately the author’s responsibility to ensure they are receiving human-narrated work. If you hire a narrator, the exposure is yours.
The royalty model changed, and the old one expires
Independently of the AI question, the economics underneath every ACX decision are being replaced this year, and the deadline is close.
Audible now offers “a 50% royalty rate for exclusive distribution and a 30% royalty rate for non-exclusive distribution”, against the legacy 40% and 25%. Enrollment opened to all creators on 26 May 2026. Then the clock: “By end of year, the legacy royalty model will be discontinued. By that time, you must either choose to enroll your titles in the new model or discontinue distribution if you prefer.”
That is 104 days from the date on this page, and two details make it a task rather than a notification. New titles are enrolled automatically; existing ones are not, so the rights holder enrols them. And enrollment “goes into effect on the 1st of the following month”, which makes November the last genuinely safe month to act.
The rate rise is real — a quarter more on exclusive, a fifth more on non-exclusive — but it is not a straight pay rise, because the base moved with it. Under the new model a member’s subscription fee, minus taxes and fees, becomes a pool called Member Value, “divided proportionally among the titles the member engaged with based on their a la carte price” before your rate is applied. Cash purchases are untouched: ACX is explicit that “our model for non-member and member cash purchases will not change”. One title now earns on two different bases depending on how each listener consumes it, and you control neither the pool nor its division.
The offsetting gain is real: the new model pays across all-you-can-listen engagement, which the legacy per-sale model did not. Whether that exceeds what the old 40% produced depends on a pool size Audible has not published.
Why 50% can pay less than 40%
Here is the comparison nobody in this category runs, and it is why the AI question and the royalty question cannot be answered separately.
Set the two Amazon-authorized options side by side. KDP pays “a 40% royalty rate for audiobooks with virtual voice” and requires a list price between $3.99 and $14.99. ACX pays 50% exclusive. On rate alone, 50 beats 40 and the AI route looks like the expensive choice.
But a rate is applied to a price, and the two programmes differ on who sets it. With virtual voice you set the list price yourself inside the band. With ACX you do not: “Audible determines your title’s price and reserves the right to adjust it at any time… Your suggested price is one factor we consider. The price at which we sell your title may not always match your suggested price.” That the ability to suggest a price ships as a new-model feature tells you what the old arrangement was.
That makes the honest comparison a break-even rather than a ranking. At the top of the virtual voice band, $14.99, a 40% royalty is $6.00 a sale and you can compute it before you publish. For ACX exclusive at 50% to match $6.00, Audible has to land on about $11.99. For non-exclusive at 30%, it has to land on about $19.99.
| Price Audible sets | ACX exclusive (50%) | ACX non-exclusive (30%) | vs virtual voice at $14.99 |
|---|---|---|---|
| $5.55 | $2.78 | $1.67 | less than half |
| $7.55 | $3.78 | $2.27 | 37% less |
| $11.99 | $6.00 | $3.60 | break-even on exclusive |
| $15.55 | $7.78 | $4.67 | 30% more on exclusive |
Those prices are not invented. They are the endpoints of Audible’s own suggested-price guidance: $5.55–$13.20 for a one to three hour book and $7.55–$15.55 for three to five hours. A short book therefore sits in a band that straddles the break-even, and where inside it you land is not your decision. Non-exclusive is starker — $19.99 is above the top of the guidance band for anything under ten hours, so for a short title the 30% rate essentially cannot match what virtual voice pays at its ceiling.
None of which makes virtual voice the better product: it is invite-only, reaches Amazon and Audible only, uses Amazon’s voice rather than yours, and produces no file you can take elsewhere. The point is narrower and more useful. The headline rates are not comparable, because only one of the two lets you set the number the rate is applied to.
Our arithmetic on published rates, not a vendor figure. Both programmes also weight membership and credit listening by the title’s price — Amazon calls its version the Audible Listener Allocation Factor — so the price you do or do not control is an input to both royalty paths, not just to cash sales.
The seven-year lock is not seven years
The most repeated warning about ACX is that exclusivity ties your book up for seven years. The term is real, and it applies to both deal types: the licence runs to “the date that is 7 years from such date (such 7 year period, the ‘Initial Distribution Period’)”, then renews automatically in one-year terms unless either party gives 60 days notice.
But the agreement contains an exit the warning almost never mentions. A rights holder may request withdrawal “at any time after the 90 day anniversary of the date that Audible first makes the Audiobook available for sale”, processed within 30 days. The practical commitment is therefore nearer 120 days than seven years — and since non-exclusive carries the same seven-year term, going non-exclusive to avoid a long contract does not achieve that. KDP Select is the closer analogue: 90 days, auto-renewing, where the binding constraint is scope rather than duration.
Where a voice clone does work
If you have already paid for Professional Voice Cloning, the clone is not wasted. It is pointed at the wrong store.
Spotify for Authors accepts it, with a disclosure it writes for you. Spotify defines digital voice narration as “using synthetic-voice technology instead of a human voice to narrate your audiobook”, asks you to tick the option at upload, then adds “a short sentence to your book’s description to let listeners know it uses digital narration”. The catch is reach: “Spotify doesn’t currently share audiobooks with digital voice narration to referral partners”, so the title stays on Spotify rather than flowing out to other retailers.
That is the third disclosure regime this site has read at source, and all three differ. Amazon wants a private declaration and shows readers nothing. YouTube requires a public label on realistic synthetic content but exempts cloning your own voice. Spotify writes the label for you and charges you distribution reach for it. There is still no portable answer to “do I have to disclose AI?”
The craft half of the old page survives intact, and one detail is neater than we realised: ElevenLabs’ input guidance for a voice clone — record “between -23dB and -18dB RMS with a true peak of -3dB”, deliver “MP3, 192kbps or higher” — is the same specification ACX sets for finished audiobook output. The recording discipline transfers exactly. Only the destination is closed.
Two corrections to our own earlier numbers. We said 30 minutes of audio, ideally 45 to 60; ElevenLabs sets 30 minutes as the minimum but recommends “closer to 2-3 hours of audio” — two to four times what we suggested. We said training takes two to four hours; ElevenLabs says “usually fine-tuning will take 3-6 hours”, sometimes up to 24. PVC needs the Creator plan or above, $22 a month with the first month advertised at half price; Free and Starter have no PVC slots. Per-hour narration costs across tools are in our AI voice generator cost per hour comparison, and the rivals in ElevenLabs alternatives for creators.
What we could not verify
Stated plainly rather than filled in with numbers we could not read at source.
- Any earnings figure under the new royalty model. Member Value depends on a pool Audible has not published and on how many titles each member engages with. No dollar estimate of what a membership listen pays appears anywhere on this page, and any calculator offering one is guessing at the same missing number.
- Whether cloned-voice submissions are rejected in practice. We have read the rule, not its enforcement. Reports of both acceptances and rejections circulate; we can measure neither and state no rate.
- Current eligibility for either AI route. Both are described as invite-only betas, and Amazon says it plans to grow the virtual voice beta over time. We could not read a published eligibility test for either, so we describe them as invite-only rather than as closed.
- The ElevenLabs distribution route, as currently constituted. ElevenLabs advertises publishing to Spotify and other retailers through a Findaway Voices partnership, and quotes keeping 100% of royalties on Spotify and 80% elsewhere. That announcement is dated February 2025 and predates Findaway’s rebrand to Voices by INaudio, which is reported rather than something we read from either company. We repeat the figures as ElevenLabs’ claims of that date and would confirm the route before planning around it.
- Non-US terms. Every figure here is USD and US-facing. Virtual voice is described as a US-marketplace beta and Voice Replicas as a US-only beta; ACX pays in USD, GBP, EUR and CAD, and we did not read the non-US price guidance.
- What ACX counts as “otherwise authorized”. The submission requirements use the phrase without defining it. We treat the two named programmes as the authorized set because they are the two Amazon documents, but ACX does not say that list is exhaustive.
FAQ
Does ACX allow AI-narrated audiobooks?
No, not unless Amazon has authorized it. ACX audio submission requirements state 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. That wording covers a clone of your own voice as much as a stock synthetic voice, because the rule turns on authorization rather than on whose voice it is. Two authorized exceptions exist, both controlled by Amazon: audiobooks with virtual voice on KDP, and the Narrator Voice Replicas beta on ACX. Neither of them accepts an audio file you produced with an outside tool.
Can I upload an ElevenLabs voice clone to ACX?
No. The two authorized AI routes both generate the voice themselves rather than accepting one you bring. ACX describes its Narrator Voice Replicas beta as a process where participants submit a sample voice recording, which ACX will turn into a high-quality replica of the participant own voice; it is invite-only, US-only and aimed at professional narrators rather than authors. KDP audiobooks with virtual voice use Amazon own synthetic voices and are also an invite-only beta. So the thing an ElevenLabs Professional Voice Clone gives you, a portable model of your voice that produces files you can take anywhere, is the one thing neither Amazon route is set up to accept.
What are the new ACX royalty rates for 2026?
Audible now pays 50% for exclusive distribution and 30% for non-exclusive, replacing the legacy 40% and 25%. Enrollment opened to all creators on 26 May 2026 and the legacy model is being discontinued by the end of 2026, at which point ACX says that to continue distributing with Audible you will need to enroll your titles in the new model. Existing titles are not converted for you; the rights holder enrolls them, and enrollment takes effect on the 1st of the following month. The rate rise is not the whole story, because under the new model membership listening is paid from a pooled Member Value rather than from your title sale price.
Does Amazon virtual voice pay more or less than ACX?
Virtual voice pays a lower rate on a price you control, which is why the comparison is not simply 40 against 50. KDP pays a 40% royalty on audiobooks with virtual voice and requires you to set a list price between $3.99 and $14.99, so at the ceiling you can calculate $6.00 a sale before you publish. ACX pays 50% exclusive, but Audible determines your title price and reserves the right to adjust it at any time, with your suggested price treated as one factor. For the 50% rate to beat $6.00 the price Audible lands on has to be about $11.99 or higher. Audible own suggested-price guidance for a three to five hour book runs from $7.55 to $15.55, a band that straddles that break-even.
Where can you publish an AI-narrated audiobook?
Spotify for Authors accepts it with disclosure. Spotify defines digital voice narration as using synthetic-voice technology instead of a human voice, asks you to tick a box at upload, and then writes the disclosure into your book description for you. The catch is reach: Spotify states it does not currently share audiobooks with digital voice narration to referral partners, so a digitally narrated title stays on Spotify rather than flowing out to other retailers. ElevenLabs also advertises distribution to Spotify and other retailers through a Findaway Voices partnership, though that announcement dates from February 2025 and predates the rebrand of Findaway to Voices by INaudio, so confirm the current route before planning around it.
What we would actually do
Decide the destination before you buy the voice. That is the inversion this page exists to make: narration is the last decision in an audiobook project, not the first, and the store you are aiming at decides whether a synthetic voice is an asset or a disqualification. If Audible is the goal, budget for a human narrator. If reach beyond Audible matters less than shipping, Spotify will take the digital narration and label it for you.
If you already have a title on ACX, the calendar item comes first: the legacy model is discontinued at the end of 2026, existing titles are not enrolled for you, and enrollment only takes effect on the 1st of the following month. And treat the 50% headline with the suspicion we applied to KDP’s 70% option — a higher share of a number somebody else sets is not automatically more money.
Related reads: KDP royalties and when 70% pays less than 35%, what an hour of AI narration costs, ElevenLabs alternatives for creators, Spotify’s podcast monetization rules, and YouTube’s AI disclosure rules.