Why One Narrator Isn’t Enough

On the page, the reader always knows who is talking: the dialogue tag says so. In audio, tags are noise. “Said Pip” after every line is exhausting to listen to, but drop it and a single voice makes three characters sound like one person talking to themselves. Distinct voices solve this: the listener identifies the speaker in the first half-second, the tags can go, and the scene moves at the speed of real conversation.

It also changes how the story is written. Once you know each character will be heard, you start giving them different rhythms, vocabulary, and sentence lengths — the same discipline covered in Building Compelling Characters, now with a practical payoff.

Step 1: Build a Voice Casting Sheet

Before generating anything, write down who speaks and how. A casting sheet is one line per character: role, a two-or-three-word voice direction, and the pitch and pace you will use. It stops you from re-deciding the villain’s voice every chapter, and it is what keeps a character consistent across a whole book.

• Cast by contrast. If the hero is high and quick, make the mentor low and slow. Two similar voices in one scene is the most common mistake.
• Give the narrator the most neutral voice in the cast. Everyone else should be heard as a departure from it.
• Limit yourself to four or five voices per story. Listeners lose track beyond that, and so will you.
• Write the direction in behaviour, not adjectives: "rises at line ends" is more useful than "cheerful".

Step 2: Split the Script by Speaker

Turn the prose into a script: every line of dialogue gets the speaker’s name in front of it, everything else belongs to the narrator. Strip the dialogue tags while you are at it — the voice is the tag now. A short passage ends up looking like this:

NARRATOR:   Pip crept between the shelves, whiskers twitching.
PIP:        Who's there?
LIBRARIAN:  Only a reader, child.
RAT KING:   And a thief, I think.
NARRATOR:   The Librarian sighed.
PIP:        I'm not afraid of you.

Step 3: Generate Each Voice

Work one character at a time, not one line at a time. Collect all of Pip’s lines for the chapter, generate them with Pip’s voice and settings from the casting sheet, export, then move to the next character. For the voices themselves we use AnyVoice, an AI voice generator with enough distinct voices to cast a full scene, plus the pitch and speed controls the casting sheet depends on. Any generator with per-voice settings will work; the discipline is the same.

Step 4: Assemble the Scene

Open a free audio editor (Audacity, or any DAW) and give each voice its own track. Place the clips in script order, leave roughly half a second of silence between speakers, and a full second at paragraph breaks. Don’t overlap lines unless the script actually calls for an interruption.

Two finishing touches make a noticeable difference. First, normalise every track to the same loudness — generators vary, and a quiet villain is not menacing. Second, add a light room reverb to all tracks together so the voices sound like they share a space instead of four separate recordings.

When to Stay With One Voice

Multi-voice narration is extra work, and it is not always the right call. Literary fiction with long interior passages, first-person stories, and anything under about five hundred words usually sounds better with a single good narrator — see our guide to turning a story into an audiobook for that workflow. Reach for a full cast when dialogue carries the scene: children’s stories, comedy, mysteries with interrogations, and anything you would describe as a “drama” rather than a “reading”.