A genuinely different voice per character
Not a pitch shift on the same read. A separate voice, cast to the part, held consistently across the book.
Audiobook Narration
AI Audiobook Narration
AI audiobook narration turns a finished manuscript into spoken chapters. The part that decides whether it works is not the voice quality. It is whether a listener can tell who is speaking.
- Use it without an account.
One block per paragraph, one voice per block. Narration takes one voice, each character takes another. Revising a paragraph re-generates that block and nothing else.
Block 1 didn’t generate.
26 words0:00
Block 2 didn’t generate.
6 words0:00
Block 3 didn’t generate.
8 words0:00
40 words · about 40 credits
A finished example
A Finished Example
The same passage twice. First with one voice throughout. Then with three voices and per-line direction.
One Voice Throughout
Narration and both characters read by the same voice.
Not playing
Who reads whatNarrator1One voiceCharacter 11One voiceCharacter 21One voice
Three speakers, the same voice each time. The names are on the page, not in the sound.
Three Voices Directed
A voice per speaker, and two lines directed by hand.
Not playing
Who reads whatNarratorNASophie ParkCharacter 1C1Abigail ReynoldsCharacter 2C2Adrian Tanaka
A voice each, and the lines marked Directed carry a written direction instead of a preset.
The passage
The same words both times. Only the voice changes.
NarratorOne voiceThe room had not changed since the last time either of them stood in it, which was the part neither of them wanted to say first.
Character 1One voice"You kept it exactly the same."
Character 2One voice"I kept it the way it was left."Directed: Under the breath, slower, weary.
NarratorOne voiceShe put her bag down by the door, where it had always gone, and neither of them mentioned that either.
Character 1One voice"You could have called. Any of the three years."
Character 2One voice"So could you."
NarratorOne voiceIt was not an argument yet. It was the shape an argument takes before anyone decides to have it.
Character 1One voice"I am not here about the house."Directed: Steadier than the line before it, holding the pause.
NarratorOne voiceAnd for the first time since she arrived, he looked directly at her.
The point of the pairing is that the second version is not “better audio”. It is comprehensible where the first is not.
The problem
The Problem Worth Naming First
The thing that decides whether an audiobook works is not how human one voice sounds. It is whether a listener can tell the speakers apart.
The failure mode is not a robotic voice.
It is a cast where everyone sounds the same, so the listener loses the thread and stops. Close your eyes during a scene with three speakers and you should still know who is talking.
Fixing it takes two things.
Control over how individual lines are delivered
Per line, not per chapter. The line that carries the scene gets directed; the rest does not have to be.
This page is about both.
This page is about how to create an audiobook that way: how to turn a book into an audiobook with a cast rather than a single reader, chapter by chapter, from a manuscript you already have.
How it works
How AI Audiobook Narration Works
Bring the manuscript in
Import from a document or a URL, in .txt, .docx or .pdf. There is no ePub import, so if your manuscript lives in an ePub, export it to Word or PDF first.
Work chapter by chapter rather than loading a whole book at once. It keeps re-generation cheap when you revise.
Your manuscript
.txt · .docx · .pdf · URL
No ePub import. Export an ePub to Word or PDF first.
Cast it
Split the chapter into blocks, then assign a voice per block. Narration takes one voice, each character takes another.
There are 1,411 voices to cast from, in three quality tiers, Emotional, Premium and Standard. For fiction the Emotional tier is usually the one worth the credits, because the whole job is delivery.
If a character needs a voice that does not exist in the catalogue, there are three ways to make one. Clone from a 10 to 30 second sample. Describe the voice you want in words and generate it. Or blend two existing voices into a third.
NANarratorEmotional
C1Character 1Emotional
C2Character 2Premium
EmotionalPremiumStandard
Cast from 1,411 voices, or make one that does not exist yet.
Direct the delivery
This is the step that separates an audiobook from a document being read aloud. Every block is directed one of two ways, and you choose which.
Pick a preset. One of 7 emotion presets: Neutral, Happy, Sad, Angry, Excited, Calm or Serious. Each one applies a matched sample of the voice, which makes it the consistent option.
Or write the direction. Leave the presets alone and a writing box appears. Describe pitch, volume, pace and texture in a few words, which is how you get a line delivered as mocking, or weary, or under the breath.
The tradeoff, stated plainly: written direction has no matched voice sample behind it, so a line you direct by hand can sound slightly less like the character you cast. On a 90,000-word book that adds up. Direct the lines that carry a scene, and let presets handle the rest.
You can also place inline sound tags and timed pauses, and set pronunciation once for invented names so chapter 2 and chapter 19 agree.
Direct this block
NeutralHappySadAngryExcitedCalmSerious✎ Write it instead
“Under the breath, slower, weary.”
A preset or a written direction: one or the other, per block.
Export chapter by chapter
Export as MP3, as a voice and music mix, or as SRT captions.
Keep chapters as separate projects. Revising chapter 7 should never mean re-generating chapters 1 through 6.
Paste A ChapterMP3chapter-07.mp3
MIXchapter-07-voice-and-music.mp3
SRTchapter-07.srt
One project per chapter. Chapter 7 re-generates on its own.
Your manuscript
.txt · .docx · .pdf · URL
No ePub import. Export an ePub to Word or PDF first.
NANarratorEmotional
C1Character 1Emotional
C2Character 2Premium
EmotionalPremiumStandard
Cast from 1,411 voices, or make one that does not exist yet.
Direct this block
NeutralHappySadAngryExcitedCalmSerious✎ Write it instead
“Under the breath, slower, weary.”
A preset or a written direction: one or the other, per block.
MP3chapter-07.mp3
MIXchapter-07-voice-and-music.mp3
SRTchapter-07.srt
One project per chapter. Chapter 7 re-generates on its own.
The honest answer
Will It Sound Like A Performance?
Three answers, and they carry the same weight. Where it holds up, where it is harder, and where a human narrator wins.
Clear, consistent, well-paced reading
Non-fiction, memoir, single-narrator work, and dialogue where the characters are distinct in the writing. Anything where the job is clear, consistent, well-paced reading.
Weight that lives in the pause
Fiction that leans on performance rather than text. If the emotional weight of a scene lives in a pause and a catch in the voice rather than in the words, you will be fighting the tool.
And it is worth saying
A single narrator who can hold a character across three hundred pages and vary them by instinct will beat this.
What this gives you instead
- Speed
- Revisability
- The ability to produce a book at all when hiring a narrator is not on the table
The honest summary: this makes an audiobook that a listener can follow and enjoy. It does not make a performance that wins awards.
Across a whole book
Long-Form Practicalities
A chapter is a demo. A book is the job. These are the four things that decide how much of it you spend twice.
Revising after you have narrated
Re-generate the affected blocks. This is the whole reason to keep one block per paragraph.
Keeping a voice consistent across a book
Cast once and reuse the project settings. Set the voice and language for each speaker at the start, rather than per chapter.
Pronunciation of invented names
Set it once per project. Fantasy and science fiction live or die on this.
Multiple languages
84 base languages, 160 including dialects. A finished narration can also be dubbed into 30+ languages.
Before you publish
Rights, Disclosure And Distribution
Rights
Selling the audio you produce
Commercial rights come with paid plans, and the free plan is for personal and evaluation use only. That covers your right to sell the audio you produce.
Full detail on pricingDisclosure and distribution
Whether a store accepts it is their policy
Whether a given store or subscription service accepts an AI-narrated audiobook is their policy, not ours, and it changes. Check your distributor's current position before you produce a full book. We make no claim either way, and you should be wary of any tool that does.
Pricing
What AI Audiobook Narration Costs
The free plan is $0 and paid plans start at $19/mo. You spend credits, and for text to speech one credit is one word. That makes a long book easy to price before you start: a 90,000-word manuscript is about 90,000 credits.
Free
Paste A Chapter$0
For hearing the cast and judging the delivery on a sample chapter.
- Use it without an account
- 1 credit = 1 word for text to speech
- Commercial rights come with paid plans
If you intend to sell it
Paid plans
See pricingfrom$19/mo
A 90,000-word book does not fit the entry plan. Plan on the tier above it.
- Commercial rights included
- Export MP3, a voice and music mix, or SRT
- Dub a finished narration into 30+ languages
How credits work
1 credit=1 word
A manuscript of90,000 words
50,000 credits: entry plan
Credits to narrate itabout 90,000
A book that size does not fit the entry plan, so plan on the tier above it.
Full detail on pricing →Regenerating a chapter spends those words again. It pays to get the casting and the direction right on a sample chapter before you run the whole book.
FAQ
Frequently Asked Questions
Yes. Assign a voice per block, so narration and each character can differ.
Yes, with 7 emotion presets and a short written instruction per block.
Commercial rights come with paid plans. Check your distributor's own policy on synthetic narration separately, because platforms differ and some require disclosure.
There is no fixed chapter length. Cost is credit based.
Re-generate that block. The rest of the chapter is untouched.
That is your distributor's policy and it changes. Check their current position before producing a full book. We make no claim about acceptance.
Most platforms now require it. Follow your distributor's rules; this page does not advise you otherwise.
About 90,000 credits, because one credit is one word for text to speech.
Import the file, split it into chapters, assign a voice to the narrator and one to each character, direct the lines that carry a scene, then export chapter by chapter.
Yes, and that is the whole job of this page. You do the casting and the direction that a narrator would otherwise do.
No. Export it to Word or PDF first.