How to make a multi-voice audiobook with AI character voices

Fiction lives in its dialogue. When one voice reads every line, the listener has to work out who is talking from the dialogue tags alone, and in a fast exchange between three people that gets tiring. A multi-voice audiobook fixes that: the narrator tells the story, and each character speaks in a voice of their own.

Hiring a cast of human narrators for this is expensive and slow to coordinate. With AI narration it becomes a casting decision you can make in an afternoon. This guide shows how character voices work in VoicedBook, how to set them up step by step, and what we have learned about making a cast that sounds like a book rather than a radio play.

Single narrator, multi-voice or full cast

These terms get mixed up, so here is how we use them.

• Single narrator. One voice reads everything, including the dialogue. This is still the most common format for audiobooks, and it suits most non-fiction.
• Multi-voice. A narrator reads the prose, and characters speak their own quoted lines in their own voices. This is what VoicedBook character voices do.
• Full cast. Often used for productions closer to audio drama, with sound design and performed scenes. VoicedBook does not add music or effects.

For most novels, multi-voice is the sweet spot. The narrator keeps the book together, and the cast makes conversations easy to follow.

How character voices work in VoicedBook

The whole feature rests on one simple marker in your manuscript: [voice: Name]. Put it in front of a paragraph, and the quoted speech after it in that paragraph is read by that character's voice. Everything outside the quotes stays with the narrator.

[voice: Elizabeth] "I am perfectly serious," she said, folding the letter. "He will not come."

In that paragraph, Elizabeth's voice reads both quoted lines. The narrator reads "she said, folding the letter". The marker itself is never spoken and is not counted as a word.

A few rules cover the rest:

• A new paragraph goes back to the narrator unless it has its own marker.
• A paragraph with no quotes at all is read entirely by the marked voice. That is handy for a letter, a diary entry or a character's inner monologue.
• [voice: narrator] hands a passage back to the narrator by hand.
• Each character gets either a studio voice or one of your cloned voices. A character you leave without a voice is read by the narrator.

Step by step: casting your book

1. Upload and review the manuscript

Upload your PDF, EPUB or Word file and check the extracted text on the Manuscript page, as you would for any book. Fix extraction mistakes first. Dialogue detection works on the text you approve, so clean text gives cleaner casting.

2. Run Find speakers

On the Voice page, open the Character voices card and run Find speakers. It reads the book, works out who says each quoted line, and inserts the markers for you. It then casts the characters: each one who speaks at least twice gets a studio voice picked by gender and by age or manner words from the text, every voice is different, and none of them is the narrator's voice. Up to 12 characters are cast automatically. Characters with only one line stay with the narrator until you choose a voice for them.

Find speakers is included in the price. Paragraphs you have already marked by hand are kept as they are.

3. Check the cast

The Character voices card lists every character it found. For each one you can change the voice, pick one of your cloned voices, or leave it empty so the narrator reads that character. Each character also has a short notes field for how they should sound, for example "older, dry, unhurried".

4. Fix individual lines

No detection is perfect, especially in long rapid exchanges where the tags drop away. Open the chapter in the Manuscript editor and use the Voice menu to insert a marker where a line is wrong. You can also type the marker yourself.

5. Audition before you narrate

Run a free audition on a dialogue-heavy chapter. Audition samples keep whole paragraphs, so you hear the cast trading lines with the narrator. If two characters sound too close, change one of them now, while it costs nothing.

6. Narrate

Start narration as usual. Each voice run is checked word by word against your text like any other passage, and anything that drifts is retaken. The finished chapters are mastered and packaged the same way as a single-narrator book.

Recasting later only touches that character

Say your beta listeners tell you the villain sounds too friendly. Change his voice, or edit his notes, and only the passages where he speaks are marked out of date. The rest of the book keeps its audio. Changing the narrator's voice is the exception, because the narrator is in every chapter.

Tips for a cast that sounds like a book

• Keep the cast small. Give voices to the characters who carry scenes. A shopkeeper with two lines often sounds more natural read by the narrator.
• Make voices easy to tell apart. Two characters who often talk to each other should differ clearly in pitch or manner. Audition their scenes together.
• Be consistent with quotes. Detection looks for quoted speech. Double quotes are the safest choice; single quotes are only used for dialogue in paragraphs that have no double quotes.
• Keep your dialogue tags. "She said" sounds right in the narrator's voice and helps the listener in long scenes. You do not need to remove them for a multi-voice book.
• Use a cloned voice where it matters. If you narrate your own memoir or novel with your cloned voice, you can keep studio voices for the other characters, or the other way round.
• Check scene breaks and flashbacks. Those are where a quoted line is most likely to be given to the wrong person. Skim those chapters in the editor after Find speakers.

What it costs

Character voices do not change the price. You pay $3 per 1,000 manuscript words for the initial narration whether one voice reads the book or ten. Markers are not counted as words, and Find speakers is included. A cloned voice in the cast follows the same account rules as a cloned narrator; the Voice page shows them.

Names in a big cast are also where AI narrators trip most often. Before you narrate, read our guide to fixing AI narration pronunciation and set up the names once.

Common questions

Do character voices cost extra?

No. The price stays $3 per 1,000 manuscript words. Voice markers are not counted as words, and Find speakers is included.

Does the narrator still read the dialogue tags?

Yes. Only the quoted speech goes to the character. Words outside the quotes, such as "she said" or "he looked away", stay with the narrator.

Can I use my own cloned voice for one character?

Yes. Any character can use a studio voice or one of your cloned voices. The same account rules apply as for a cloned narrator, and the Voice page shows them.

What happens if I change a character's voice after narration?

Only the passages where that character speaks are marked out of date and re-narrated. The rest of the book keeps its audio.

How many characters can Find speakers cast?

It gives a studio voice to up to 12 characters who have at least two lines. Characters with fewer lines stay with the narrator until you pick a voice for them.