How to Create Audiobook Character Voices with AI

How AI gives each character their own voice, how to cast your book so it sounds right, and what each tool and store allows.

Published · 6 min read
In this article

A good narrator makes every character sound different, and that’s one of the hardest things to pay for with human narration. A cast of narrators costs more again. AI changes that: it can give each character their own voice for no extra cost. To create character voices with AI, use a tool that finds each line of dialogue and who says it, give every main character a voice that suits them, let the narrator read everything else, and listen back to check each line went to the right voice.

This guide covers the different styles of multi-voice audiobook, how AI does it, how to cast your book well, and what each tool and store allows.

Single narrator, dual, duet or full cast?

Human audiobooks come in a few styles, and AI can do all of them.

StyleHow it worksBest for
Single narratorOne voice reads everything, changing tone for charactersMost nonfiction, literary fiction, first-person stories
Dual narrationTwo narrators take turns by chapter or point of viewRomance and stories with two narrators
Duet narrationTwo narrators voice the main characters throughoutRomance with two leads
Full castA different voice for every main characterFantasy, thrillers, children’s books, dramatized stories

Dual and duet narration are growing fast, especially in romance. ACX says dual narration “pairs two narrators who trade off by chapter or point of view”, while duet narrators voice the main characters all the way through (ACX’s guide to dual and duet narration).

A full cast is what GraphicAudio is known for: actors for each character, plus music and sound effects. If you want to go that far, see our guide to making an audiobook with music and sound effects.

With human narrators, every extra voice costs more. ACX contracts also support only one producer, so extra narrators are hired outside ACX.

With AI, extra voices cost nothing extra, so the choice comes down to what suits your book.

How AI character voices work

Every multi-voice AI tool does the same three things:

  1. Find the dialogue. The tool goes through your manuscript and marks every line spoken by a character.
  2. Work out who says it. It reads the dialogue tags (“said Mara”) and the context to decide which character is speaking.
  3. Give each speaker a voice. You pick a voice for each character, or let the tool suggest a cast. Everything else is read by the narrator.

Google describes the last step plainly: “Text that’s not assigned to a character is spoken by the default narrator.”

ElevenLabs’ Characters step, listing each character in Pride and Prejudice with a voice
Multi-voice in practice: ElevenLabs finds the characters and gives each a voice.

How to create character voices, step by step

1. Make your dialogue easy to follow

AI works out who’s speaking from your text, so help it. ElevenLabs says “explicit dialogue tags yield more predictable results than relying solely on context”.

  • Add a tag (“said Finn”) every few lines in long conversations.
  • Start a new paragraph for each new speaker.
  • Watch out for dialogue split by action: “Wait,” Mara said, grabbing his arm. “Listen.” Both halves belong to Mara.

Our guide to preparing your manuscript for AI narration covers the rest of the prep.

2. Find your characters

Let your tool detect the characters, then check its list:

  • Merge duplicates. “Elizabeth”, “Lizzy” and “Miss Bennet” may be one person.
  • Split mix-ups. Two characters with the same surname can be merged by mistake.
  • Look for strays. A word in quotes, like a sign or a nickname, may show up as a character.

3. Cast the main characters

Choose a voice for each character with a real part to play. Good casting comes down to a few rules:

  • Contrast. Characters who talk to each other should sound clearly different: vary gender, age, pitch and pace.
  • Fit. Match the voice to who the character is. A gruff old fisherman and a nervous teenager need very different voices.
  • Accent. Match the setting where it matters, but don’t mix accents for no reason. Use accents respectfully, and don’t make one a joke.
  • Test with real lines. Hear each voice read the character’s own dialogue, not a sample sentence.

4. Choose your narrator

The narrator reads most of the book, so choose them first and choose carefully. They should be easy to listen to for hours and sound different from your main characters. In a first-person book, consider using the narrating character’s voice for the narration too.

5. Leave minor characters to the narrator

A shopkeeper with one line doesn’t need their own voice. Too many voices make a book sound crowded and make it harder to follow. A good rule: give a voice to any character with more than a handful of lines, and let the narrator read the rest.

6. Listen back

AI will get some speakers wrong, usually in long back-and-forth conversations or where a line has no tag. Listen to every chapter with the text in front of you, and reassign any line that went to the wrong voice.

Doing it in Audiobook Generator

Audiobook Generator finds every line of dialogue and who says it, then lists your characters with how many lines each has. Choose a voice for each, or press Cast for me to get a suggested cast. Minor characters, or everyone, can be read by the narrator instead.

Cast14 characters found Cast for me
CharacterLinesVoice
N Narrator2,140Maisie
E Elizabeth512Lily
MB Mrs. Bennet236Alice
D Mr. Darcy198Daniel
B Mr. Bennet121George
J Jane104Sarah
Every character found in the dialogue, with a voice for each. Press play to hear a voice read your own words.

The same voice reads a character in every chapter, so Elizabeth sounds like Elizabeth from chapter 1 to chapter 61. If you have a series, use the same voices in every book. It’s one of the main reasons to use an AI audiobook generator with a cast rather than a single voice.

What other tools support

ToolCharacter voices?Notes
ElevenLabs StudioYesAssign voices to characters and paragraphs
Google Play auto-narrationYes, English onlyDetects characters; you can’t undo detection
Amazon Virtual VoicePer chapter onlyA different voice for each chapter, not each character
Apple digital narrationNoOne voice, chosen by genre
Spotify’s creation toolsNoOne digital voice

Store rules for multi-voice AI audiobooks

The stores that accept AI narration ask you to label it, however many voices you use:

  • Spotify: tick the digital voice narration box.
  • Google Play: set each narrator to “Synthesized voice”.
  • Kobo: list the narrator as a synthesized voice.
  • Apple Books: say it’s AI in the narrator field and description. Apple also asks for one name per narrator field and doesn’t accept generic names like “Narrator 1”.

Our guide to where you can publish AI-narrated audiobooks has the full list.

Do listeners like AI character voices?

Early research is encouraging. In a study by Edison Research for Spoken, an AI audiobook company, fiction listeners rated a multi-voice AI audiobook more favourably than a human-narrated one (61% against 53%), and were about as likely to buy it (details). The study was commissioned by a company that sells AI narration, so treat it as a promising sign rather than the final word.

FAQ

Can AI do different voices for characters?

Yes. Multi-voice AI tools find who says each line and read it in that character’s voice, with a narrator for the rest.

How many voices should an audiobook have?

A narrator plus one voice for each main character is usually right. Leave minor characters to the narrator so the book doesn’t sound crowded.

What is a full cast audiobook?

An audiobook with a different voice for each main character, like a radio play. Some full-cast productions add music and sound effects too.

What’s the difference between dual and duet narration?

In dual narration, two narrators take turns by chapter or point of view. In duet narration, two narrators voice the main characters throughout, even in the same scene.

Does a multi-voice AI audiobook need a special label?

No special label, just the usual AI disclosure each store asks for, such as “Synthesized voice” on Google Play.

Ready to cast your book?

Upload your manuscript, see who says what, and hear your characters in their own voices. Your first chapter is free.

Share this article

Philip Ghezelbash is the founder of Audiobook Generator and the author of The Stoic Body, which has sold over 10,000 copies on Amazon. He writes about making, publishing and selling audiobooks.

Your book deserves to be heard.

Narrate your first chapter free. No card needed.