How to make an audiobook of your book

What an audiobook production is

An audiobook is your book read aloud, recorded, cleaned up, and delivered as audio files that meet a store’s technical rules. Every production has the same six parts.

• The manuscript. The text as it will be read.
• The narrator. You, a hired narrator, actors each reading a part, or generated voices.
• The recording. Chapter by chapter.
• The editing. The mistakes and noises come out and the right take goes on every line.
• The finishing. The levels are brought to the store’s numbers, the credits and the retail sample are added, and the files are written.
• The distribution. The upload with the cover and the metadata, and the store’s review.

What it costs when hired out. ACX, Audible’s production marketplace, publishes planning figures: about $200 per finished hour for narration and another $200 for post-production, the editing and the mastering. ACX counts about 9,300 words to a finished hour, so an 80,000-word novel is about 8.6 hours and the two together come to about $3,400. The figures are from ACX’s guide to paying for an audiobook, first posted in 2013 and still published by ACX, read on October 16, 2026.

What it costs to do it yourself. Your time, first. Expect two to three hours at the microphone for every finished hour, plus the editing and the proof listening. Then the equipment, which the guides below cover, and the software.

Prepare the manuscript

A clean file. Export the final text as a plain text file (.txt) or a Word file (.docx), without the page numbers, running headers, footnote markers and layout of the printed book.

Chapter headings the software can find. Every chapter starts with a heading on its own line: “Chapter 1”, “Chapter One”, “Prologue”. Recording software splits the book at these headings, and the stores want one file per chapter.

Front and back matter decided. The title page becomes the opening credits. Narrators read the dedication and the epigraph. They skip the copyright page, the table of contents and the list of the author’s other books.

Pronunciations noted. List every invented name, place name, foreign word and term of art with how it is said, and give the list to the narrator before recording starts. A name said two ways in one book is one of the things a store’s reviewer listens for.

A character list for a novel with dialogue. Every speaking character, with a line on who they are and how they sound. A narrator uses it to keep the voices apart; actors need it; for generated voices it is the casting sheet.

Choose how it will be read

The reader decides the cost, the time, the control you keep, and which stores will take the result.

You read it yourself. It costs your time and a recording setup. You keep control of every line, and the book is read by the person who knows what each sentence means. Every store accepts a human reading.

A hired narrator. It costs money, at the planning figures above or more, and less of your time. The post-production figure above buys edited, finished files. Every store accepts it.

Actors for the parts. Each character read by a different person. It is the most work to organize, and the recordings have to match in sound. Every store accepts it.

Generated voices. A text-to-speech service reads the text in a synthetic voice. It costs little and is fast. What you give up is the performance. The stores differ. As of October 2026:

• Audible, through ACX: not accepted. ACX’s Audio Submission Requirements say the audiobook must be narrated by a human unless otherwise authorized, and that “unauthorized text-to-speech, AI or automated recordings” are prohibited (read October 16, 2026). Amazon’s audiobooks with virtual voice are a separate beta: KDP generates the narration from an eligible Kindle eBook, labels it, and pays a 40% royalty on single sales (read October 16, 2026).
• Spotify, through Spotify for Authors: accepted with disclosure. You tick “This audiobook uses digital voice narration” at upload and Spotify adds a sentence to the description (Spotify’s page, read October 16, 2026).
• Google Play Books: Google’s pages describe its own auto-narrated audiobooks beta, generated with Google’s voices from an ebook already on sale there (read October 16, 2026). Whether it takes audio made elsewhere is a question for the store.
• Apple Books: Apple’s page on digital narration describes Apple’s own program and does not say whether audio made elsewhere is accepted (read October 16, 2026).

A mix. A recording you made and then converted into another voice is a different thing from a generated reading, since the timing and the emphasis are yours. The export says which lines are which.

Set up to record

A home setup is a microphone, an interface, headphones and the room. Each has its own guide on this site, with the picks and the prices.

The microphone. One voice, close to the microphone, in a room that is not a studio. The job favors a microphone that hears the voice and little of the room. Keep it a hand’s width from your mouth and a little to one side. The microphone guide has a pick in every price tier.

The interface. It turns an XLR microphone’s signal into something the computer can record. A USB microphone does not need one. The interface guide covers what a narrator needs.

The headphones. Closed-back, so that what you hear does not leak into the microphone, and detailed enough to hear a click while editing. The headphones guide has picks under $75, $150 and $250.

The room. The room decides more than the microphone does. Reflections and noise are the two problems, and neither comes out in editing. The room treatment guide covers what works at each budget.

A test recording. Record ten seconds of silence, a paragraph at your normal level, and ten seconds of silence again. Then listen on the headphones and watch the meter.

• Noise floor. A hiss or a fan you can hear in the silence will be in every pause of the book. Turn it off, or move.
• Room echo. Clap once. If the clap rings, your voice rings the same way.
• Plosives. Read a sentence full of P and B sounds. A thump on each means the microphone is too close or straight in front of your mouth.
• Level. The peaks sit between -12 dB and -6 dB, with no red. Too low and raising it later raises the noise too; too high and the loud words distort, which cannot be repaired.

Record

The first session decides whether there will be a second one, so plan it to end well. Read a practice paragraph and listen back. Record the first scene, make one correction when you stumble, and play the scene back from the start.

Line by line or straight through. Reading line by line gives a clean take for each line, and a correction is one line read again. Reading straight through keeps the performance whole; a correction is made by going back to the start of the sentence and reading on, and the editing cuts the false start out. Both are divided into the manuscript’s lines by the words that were read, on your own computer, with no service or key. A recording made elsewhere can also be split at its pauses.

Pickups and takes. A pickup is a line recorded again later to replace one that went wrong. Mark the line and move on. Record the pickups in a batch, in the same place with the same setup, and they match. Keep every take until the chapter is finished and choose while editing, not while reading.

Pace. Most narration runs between 150 and 170 words a minute. Check the pace at the start of each session against a passage recorded earlier. The stores’ reviewers listen for consistency across the whole book.

See it done. In the demonstration videos, Part 2 records a line, sets the levels, reads a chapter straight through from the prompter and restarts a line. Part 3 builds the cast and gives every character a voice. Part 4 gives a character to an actor and records a part.

Edit and proof

Editing is where a recording becomes a book. Do it chapter by chapter, soon after each is recorded.

Listen back against the text. Play the chapter with the manuscript in front of you and follow every word. Mark a word that is wrong or missing, and mark a reading that does not say what the sentence means. Then fix.

Fix a line without re-recording the chapter. Every fix is to one line: swap in another take, cut a false start, or record a pickup of that sentence and put it in place.

Room tone. The sound of your room with nobody speaking, and it is not silence. When a noise is cut out of a pause, the gap is filled with room tone, or the listener hears the room drop out.

Breaths and clicks. Breaths are part of a reading. Leave the ones that belong to the phrasing, and lower or remove the loud gasps and the ones at the very start and end of a line. Mouth clicks are removed or lowered; plosive thumps are reduced.

The pauses between lines. A pause after a sentence, a longer one after a paragraph, a longer one still at a scene break. Set the three lengths once for the whole book.

Proof listening as a step of its own. When the editing is done, listen to the whole book once more, start to finish, as a listener would. While editing you hear lines. On the proof listen you hear the book, and catch a name said two ways three chapters apart or a chapter quieter than its neighbors. In the demonstration videos, Part 4 proof listens a chapter and flags the pickups, and Part 7 takes a chapter through finishing and export.

Finish

Finishing turns edited chapters into files a store will take. It is mostly measurement.

Levels and loudness. Every chapter is brought to the same average loudness, and the peaks are held under a ceiling. The listener does not reach for the volume between chapters, and nothing distorts when the store converts the files. Loudness is the average level of the file; the peak is its loudest moment; the noise floor is the level of the pauses.

The stores’ specifications, in plain words. Audible’s ACX publishes the most detailed set, and most other stores accept files made to it. It asks for an average level inside a range, peaks under a ceiling and a noise floor under a limit. It also names an MP3 format, room tone at each end of a file, a longest file length and one chapter per file. The numbers are in the ACX requirements guide.

File formats. ACX takes MP3 files at a constant bit rate, one per chapter. Some other stores take a single M4B file with the chapters marked inside it and the cover art embedded.

The opening and closing credits. The opening credits state the title, the author and the narrator, must match the cover and the metadata, and are their own file. The closing credits say that the book has ended, in ACX’s recommended form: “You have been listening to”, the title, the author, the narrator, “The End”.

The retail sample. The excerpt a customer plays before buying. ACX asks for five minutes or less, from the audiobook itself, with no explicit material.

What a passing measurement means. Software measures the loudness, the peaks, the noise floor and the format and tells you whether each chapter meets the published numbers. Meeting the measurements is not the same as being accepted by a store. ACX’s reviewers also listen to the files and check the rights to the text.

Distribute

The book goes on sale through a store or a distributor, and the choice between them is mostly the choice between exclusive and wide. The royalty figures here are as the companies published them on the dates given.

ACX, for Audible, Amazon and Apple Books. You upload the files, the cover and the sample, ACX reviews them, and the book goes on sale on those three stores. Exclusive means those stores and nowhere else for the term of the agreement, at the higher rate; non-exclusive means you can sell elsewhere too, at the lower rate. ACX’s page How royalties work, dated July 31, 2026 and read on October 16, 2026, gives the rates. Under Audible’s new royalty model, open to all ACX creators from May 26, 2026, exclusive pays 50% and non-exclusive 30%. Under the legacy model, 40% and 25%, which titles not yet enrolled keep until the end of 2026. The royalty is a share of what Audible receives, not of the list price.

Voices by INaudio, formerly Findaway Voices. A distributor. It sends the book to many stores and library services at once, and it is the common way to go wide. Its page How much will I make per sale?, read on October 16, 2026, says the author keeps 80% of all royalties Voices by INaudio receives. That is 80% of what the stores pass on, not of the list price.

Author’s Republic. Another wide distributor. Its page 80% Royalty Rate, read on October 16, 2026, announced 80% of royalties for audiobooks newly distributed between May 1, 2024 and February 28, 2025. An editor’s note on the same page says the rate has since been made permanent for newly submitted audiobooks. The 80% is again of what the distributor receives.

Direct sales, and the choice. You can also sell the audiobook from your own website through a shop service that delivers the files, with no store review and no store audience. An exclusive ACX agreement rules it out. Exclusive gives the higher rate from the largest audiobook store and nothing from anywhere else. Wide gives a lower rate per sale on Audible and access to every other store and to libraries. Which earns more depends on where your readers buy.

What the store asks for at upload. The audio files, one per chapter, with the credits and the retail sample as separate files. The cover art. For ACX it must be square, at least 2400 by 2400 pixels, JPG, PNG or TIF, RGB, no more than 8 MB. The title and the author’s name must be visible (ACX’s cover art requirements, read October 16, 2026). The metadata: title, author, narrator, publisher, description, categories, language, and whether the narration is generated where the store asks. The credits in the audio, the cover and the metadata must agree.

Where StellAurator fits

StellAurator is a Windows program for steps 2 through 7 above. It works from the manuscript. It does not distribute.

The manuscript, with every recording on its line. You import the text file or the Word file, or paste the text. The chapters are found at their headings and each paragraph is split into lines. In a novel every line gets a speaker, which you review. Every recording lands on its line with its waveform, and every line keeps every take until you delete one.

Recording. Record one line, or read a chapter straight through from the on-screen script. When you stop, the reading is divided into the manuscript’s lines by the words you read, on your own computer. After each recording the clicks, plosives, breaths, hum and background noise are marked under the take with a suggested fix for each. Nothing is applied until you press Apply, and the original recording is kept.

Record a part, for actors. You give a character to an actor, who receives a free copy of the program showing only their lines. Their recordings come back onto their lines, where you accept them or mark them for another take.

Proof listen. A chapter plays as it will export, including the pauses, with a tick on each line as it is heard. Pressing F marks a line for a pickup without stopping; the marked lines are collected in Pickups.

The Finish checks. Choose the output, Audiobook master for Audible’s ACX among them, and press the five steps in order: Diagnose, Plan, Apply, Check, Export. The report gives the loudness, the peak and the noise floor of every chapter against ACX’s published numbers, with a pass or fail on each. The store decides what it accepts.

Export. The export writes one MP3 per chapter in the form ACX requires, or one M4B with chapters for other stores. You choose which take each line exports, the locked-in one or the newest, or leave a line out. A file that contains a generated voice is marked as containing AI-generated narration, and a file with a reading you recorded and converted is marked as a speech-to-speech modified human performance.

The practice book and the trial. A practice book comes with the program, for trying every step before your own book. The studio is free for seven days, with everything except exporting the finished audiobook; exporting needs a plan, and the prices are on the pricing page. StellAurator runs on 64-bit Windows 10 and 11 only. A Mac version and phone apps are on the way, and you can ask to be told when they arrive.