AI Audiobook Translation: What Is Documented, What Is Not

We found no documented self-publisher route that takes your book from AI translation to a store-ready audiobook in one step, in the pages we read. Kindle Translate says it cannot create audiobooks, Audible's AI translation beta is for select publishers, and the ElevenLabs dubbing page we read does not describe an audiobook workflow. We also found no ACX or KDP rule on AI-translated audiobooks for self-publishers. This page sets out what is documented and what is not. This is general information, not legal advice.

Two ways to make a translated audiobook

A translated audiobook is made either from the translated text or from the finished audio, and the rules you need to read depend on which. Audible's May 13, 2025 release names the same two pathways for its publisher beta: text-to-text translation followed by narration, and speech-to-speech translation.

For most self-publishers the text route gives more control, because a bilingual reader can correct the words before you spend anything on audio. That is our judgment, not a measured result.

What each route documents

As of 2026-11-08, these are the routes named in primary sources and their limits.

ElevenLabs dubbing: the documented limits

ElevenLabs' dubbing documentation lists, as read on 2026-11-08:

• Version 2: up to 1 GB and 180 minutes per file in the app, and 3 GB per file through the API.
• Version 1 (Dubbing Studio): 1 GB and 45 minutes, and in maintenance mode.
• Languages: the version 2 table lists 94 language codes, 14 of them regional variants (the page also says "90+ languages").
• Concurrency: three concurrent jobs on self-serve plans, and all plans can create and download dubs.

The page does not describe dubbing as an audiobook workflow. Our own arithmetic shows the scale problem: a 90,000-word book is about 580 minutes of audio at ACX's 2013 figure of 9,300 words per hour. Under version 2's limit, 580 / 180 means four files. Under version 1's 45 minutes, 580 / 45 means 13. We have not established whether dubbing a finished audiobook, rather than translating text and narrating it, is sensible, and we make no claim either way. For tool detail see the ElevenLabs review. For production cost, use the cost calculator with the character count of the translated text, which can differ from your original.

Languages by route

Translation only helps if a route can narrate the target language, and the routes differ sharply.

Read by target language, the same facts look like this. "Listed" means the source names the language and says nothing about how well it performs or whether a store will take the result.

Spotify's tools are left out because they are English only. ElevenLabs is left out because the counts above are model totals, so check your language against the model you plan to use.

Having a model that speaks a language is not the same as a store that accepts it. Check the target store separately in the cross-store comparison, and see Google Play auto-narration and Spotify's creation tools for their rules.

Before you spend money: a working order

Do these in order, because each step can end the project cheaply. The sequence is our suggestion.

1. Confirm you hold the rights. Translation and audio rights are separate from the right to publish your book in English. If you licensed either to a publisher, you may not have them. YouTube's copyright help, for one, lists translations among derivative works the copyright owner controls, and rules vary by country.
2. Pick the language from the table. Choose one where a route you can actually use is listed, not one a model merely supports.
3. Ask the store in writing whether it takes an AI-translated, AI-narrated edition in that language, and keep the reply.
4. Translate a sample first and have a fluent reader mark it up. Character voice, idiom and names are what you are judging.
5. Narrate only the sample, and have the same reader listen for pronunciation, stress and names. Fix these before you generate the book.
6. Record what was AI for each stage. Translation and narration are separate facts, and a store or KDP may ask about either. Whether AI-translated text can be protected is a separate question, covered in AI and copyright for authors.

Skip this if you cannot pay for or find a fluent reader. Without one, you have no way to know what your book now says.

What we could not find

• No ACX or KDP rule specifically about AI-translated audiobooks for self-publishers.
• No statement on whether ACX treats a translated book differently. ACX still requires human narration unless authorized, per its submission requirements. See ACX and Audible.
• Whether KDP Virtual Voice titles can be AI translations, the public KDP pages we read do not say.
• Whether Audible's translation beta opened, widened or changed after the 2025 release.
• Which languages Voices by INaudio or Kobo accept for AI-narrated titles. The pages we could read do not say.
• Any measure of how accurate AI translation is for book-length fiction or nonfiction. We have not read translated text.

Silence on a point is neither permission nor prohibition. Before you spend money, ask each store, and treat translation quality as a separate question for a human reviewer who reads the target language.

For the rest of the AI narration guides, return to the AI audiobooks hub.

Last verified 2026-11-08. Rules and prices change: check the linked primary page before you rely on this.