How to Turn an Ebook Into an Audiobook

The five steps in the video

1. 1

   Open Export on your finished book

   With the book finished, open Export from the top right of the editor.

2. 2

   Switch to the Audio tab

   Choose Audio, then click Create audiobook. Narration is generated chapter by chapter.

3. 3

   Listen back and choose your voice

   Play the finished audiobook in the studio. To change narrator, click Re-record in a different voice and pick another one.

4. 4

   Download the MP3

   Download the complete audiobook as a single MP3 for personal use, podcast feeds or your own site.

5. 5

   Export per-chapter files for distribution

   Open the distribution panel and download the chapter zip: one MP3 per chapter at 44.1 kHz, 192 kbps CBR, mono, loudness-normalised, which is the format audiobook retailers expect.

Full transcript

Where an AI-narrated audiobook can actually go

Platform rules change. Check each retailer's current disclosure policy before you upload.

Choosing a narrator voice

The voice is the single decision that most changes how a finished audiobook lands, and it is worth spending five minutes on before you record the whole book. In the video the narrator is Tatyana, but the audiobook studio offers a full range, and you can re-record at any point without touching the manuscript.

Three things actually matter when you pick one. Match the voice to the book, not to your own taste. A warm, unhurried narrator suits memoir, self-help and devotional writing. A crisper, faster read suits business and how-to, where the listener is often multitasking. Fiction is the hardest case: a narrator who sounds neutral across dialogue will flatten a story that depends on character, which matters most for novels and romance.

Listen to a full chapter, not a sample line. Any voice sounds convincing for ten seconds. The tics that irritate listeners, an odd emphasis pattern, a breath that lands in the same place every sentence, only surface across several minutes. Record one chapter, play it while doing something else, and see whether you stop noticing the voice or start noticing it more.

Check how it handles your specific vocabulary. Names, places, technical terms and any invented words in fiction are where synthetic narration is weakest. If your book leans on unusual proper nouns, audition on the chapter that contains the most of them rather than on chapter one. Fiction writers can check names against their story bible first. Comparisons of the underlying narration options are covered in our guide to AI voice generators for audiobooks.

What makes audiobook audio pass quality control

Every audiobook retailer runs submitted files through an automated check before a human ever hears them, and the same handful of numbers decides the outcome almost everywhere. Understanding them is useful even when the export handles it for you, because it explains why the distribution files differ from the plain MP3 you get from the audiobook generator.

Loudness has to sit in a narrow band. The standard is an RMS level between -23 and -18 dBFS. Too quiet and listeners reach for the volume on every chapter. Too loud and the file distorts on cheap earphones. Retailers reject on either side of the window rather than fixing it for you, which is why the distribution export normalises loudness across the whole book instead of leaving each chapter at whatever level it was generated.

Peaks must stay under -3 dBFS. This is headroom, not volume. It leaves room for the encoder to do its work without clipping, and it is the check most home recordings fail.

The noise floor has to be below -60 dBFS. For a human recording this means a quiet room and decent equipment. Synthetic narration has an advantage here, since there is no room noise to remove in the first place, which is one reason AI narration passes automated checks so consistently.

Each file needs a moment of silence at both ends. Retailers expect roughly one to five seconds of clean space at the start and end of every file, so chapters do not begin the instant the listener taps play and do not cut off mid breath. Files that start flush on the first word read as abrupt and can fail review.

And the format itself is fixed: MP3 at 44.1 kHz, 192 kbps constant bit rate, mono, with every file in the book using the same channel format and none running longer than 120 minutes. The distribution zip is produced to exactly this specification, so the files upload without re-encoding.

The single MP3 and the chapter zip do different jobs

The studio gives you two downloads and it is easy to grab the wrong one. The single MP3 is the whole book in one file. It is the right choice for listening back yourself, for sending to a beta reader, for a podcast feed, or for selling directly from your own site where you control the player, alongside the ebook editions you already sell. It is the wrong choice for a retailer.

The chapter zip contains one file per chapter, numbered so the order is unambiguous. Retailers use those file boundaries as the chapter markers in their apps, which is how skipping forward moves a listener one chapter rather than thirty seconds. Upload a single long file and a listener has no way to navigate a ten-hour book, which is why most platforms simply reject it.

If you edit the book after recording, the studio flags which chapters have changed and prompts a re-record. It is worth doing before you distribute, because a mismatch between the text edition and the audio edition is the sort of thing reviewers notice and mention, and it undermines the launch checklist work you did on the book.

Where AI narration works well, and where it struggles

Being straight about this saves you a wasted afternoon. Synthetic narration is not uniformly good or bad across book types. It is very good at some things and noticeably weaker at others, and the difference is predictable enough that you can decide before you record.

It works well for informational non-fiction. How-to guides, business books, reference material, study guides and most self-help read cleanly. The listener wants the information, the sentences are declarative, and there is no character work to do. A steady, clear voice is genuinely the right delivery for this material, and many listeners will not think about the narrator at all, which is the goal. If your book came out of the book writer as structured non-fiction, it will almost certainly narrate well.

It struggles with heavy dialogue. Fiction that leans on distinct character voices is the hardest case. A human narrator shifts register between speakers, holds a pause for effect and lets a line land. Synthetic narration reads dialogue accurately without performing it, so a scene that depends on delivery flattens out. This does not make an AI-narrated novel unlistenable, and plenty of readers happily listen to one, but you should go in knowing the trade rather than discovering it at chapter twelve.

It struggles with anything unusual on the page. Tables, code blocks, long numeric sequences, footnotes and heavy formatting were written to be seen, not heard. A recipe list of quantities reads as a stream of numbers. If your book is dense with these, consider whether the audiobook edition needs a lightly adapted text rather than a straight reading of the print version.

Length is on your side. The traditional barrier to an audiobook was the per-finished-hour cost of a narrator, which meant long books were the most expensive to produce and short ones were rarely worth doing at all. That calculation does not apply here, so a short guide or a lead magnet can have an audio edition just as easily as a full-length work.

Why audiobook submissions get rejected

Most rejections are not about the writing. They are about the files, and they cluster into a small number of causes that are all avoidable if you know them in advance.

Undisclosed AI narration. The single fastest way to get a title pulled. Every platform that accepts synthetic narration asks you to declare it, and the declaration costs you nothing. Failing to tick the box is treated as misrepresentation rather than an oversight, and the penalty is disproportionate to how easy it was to avoid.

Submitting to a platform that does not accept AI voices at all. This is the ACX case, and it accounts for a lot of wasted effort. No amount of audio quality changes the answer, because the rule is about who narrated, not how it sounds. Check the narration policy before you prepare files, not after.

Inconsistent loudness between chapters. If chapters were recorded at different times or with different settings, the listener adjusts the volume every time a chapter turns. Automated checks catch this, and it is why the distribution export normalises across the whole book rather than per file.

One long file instead of chapter files. Covered above, but it remains one of the most common first-time mistakes. Download the right export.

Metadata that does not match. The title, author name and any spoken credits need to agree with what you typed into the platform's submission form. A book called one thing on the cover, another in the metadata and a third in the opening credits will be held for review. If you generated the cover with the cover generator, check the exact wording on it against your submission before you upload.

Pricing and royalties, briefly

Audiobook pricing works differently from ebook pricing. Listeners are used to audiobooks costing more than the ebook of the same title, because traditionally an audiobook carried studio and narrator costs that a text edition did not. Retailers often set or constrain the price themselves based on runtime, so you may have less control than you are used to from ebook publishing.

Royalty rates vary by platform and by whether you distribute directly or through an aggregator, and they change often enough that any specific number printed on a page like this would be out of date within months. Check the current rate card on each platform before you commit, and read the exclusivity terms carefully: some programmes offer a higher rate in exchange for not selling the same audiobook elsewhere, which is the same trade-off KDP Select asks of ebooks.

The economics of AI narration are simply different from traditional production. There is no per-finished-hour narrator fee and no studio time, so the calculation that stops most self-published authors from making an audiobook at all does not apply in the same way. What has not changed is that an audiobook only sells if the book does, so the writing and the cover still carry the work.

Frequently asked questions

Keep reading