How to Do Character Voices in an Audiobook, by Voice or with AI

Do you need character voices at all?

Before working out how to do character voices, decide whether the book needs them. Not always: many literary novels and most nonfiction are narrated with light shading, and a strong narrator can carry dialogue with attitude and timing alone. Genre fiction with big casts (fantasy, romance, thrillers) is where listeners expect distinct voices, and where confusion costs you. Either way, the dialogue tags in the text do half the work: “said Mara” is not a weakness in audio.

Choose a level and hold it: a shaded style where characters are suggested, or a voiced style where each has a clear identity. Mixing the two, a few big voices with everyone else flat, is what sounds uneven. The trade-offs are laid out in full cast vs single narrator.

Five levers: pitch, pace, placement, rhythm, vocabulary

Most advice on how to do voices for audiobooks starts with pitch, the obvious lever and the most overused. Stay within a comfortable range around your own voice, with the narrator in the middle. Pace changes more than pitch: an anxious teenager clips and rushes, a tired detective drags. Placement (chest, mouth, nose, throat) changes the color of a voice without moving its pitch, and it is how one narrator voices two men of the same age.

Rhythm and vocabulary come from the page. Find each character's sentence shape and pet words, then lean on them: the soldier who never uses contractions, the aunt who ends every complaint with a question. Use one or two levers per character, not five. The clearest casts are built from a few strong contrasts, especially between characters who share scenes, and that is the real answer to how to make character voices distinct.

Keep voices consistent: a character bible

Over a ten-hour book, voices drift. Keep one line of notes per character (“Mara: low, quick, chest, drops her g's”) and record a ten-second reference clip of each the first time they speak. Replay the clips before every session and before any scene where a minor character returns after a long gap.

Note where each character appears, too. A side character who speaks in chapter 2 and again in chapter 27 is the one most likely to come back sounding different. For a series, the bible carries straight into book two.

Opposite-gender and child voices without caricature

Across gender, suggest rather than imitate. A male narrator voicing a woman rarely needs falsetto: a lighter placement, softer onsets and her own rhythm do more, and the tags already tell the listener who she is. The same works the other way; a woman voicing a gruff man can drop into chest resonance and slow down rather than force the pitch.

Children are the hardest. Skip the cartoon squeak and use a brighter, forward placement, simpler phrasing, quick shifts of feeling and plenty of curiosity in the inflection. Keep the volume natural: a child shouting at full adult power breaks the illusion.

Accents: when to use them and when not

Use an accent when the story depends on it, such as a newcomer introduced as one or a plot point about where someone is from, and only if you can hold it for the whole book. A light, consistent accent beats a strong one that slips. When unsure, carry origin through rhythm and word choice, and never play an accent for laughs at a real community's expense.

Study real speakers, not other actors' versions: interviews and recordings of native speakers make better references than films. Listeners forgive a neutral voice far sooner than a wobbly one.

Doing it with AI: casting archetypes instead of impressions

Knowing how to do character voices still pays off with AI, because the levers turn into choices. Pitch, placement and age become which voice you cast: a deep, gravelly older voice for the mentor, a light, bright one for the youngest sibling. Pace and attitude become directions in brackets, such as [quick, breathless] or [dry, unhurried]. Rhythm and vocabulary stay where they were, in the text.

FullCastBook reads your manuscript, finds every speaking character and suggests a distinct voice for each, plus a narrator. You can swap any voice, reassign any line and keep the cast across a series, which does the character bible's job for you. Give the most contrasting voices to people who share scenes; AI voices for audiobooks covers choosing them and AI voice emotion tags covers directing them.

One scene, two ways (sample)

From chapter 1 of Jane Austen's Pride and Prejudice (1813, public domain). Mrs. Bennet: “My dear Mr. Bennet, have you heard that Netherfield Park is let at last?” Narrator: Mr. Bennet replied that he had not. Mrs. Bennet: “Do you not want to know who has taken it?” Mr. Bennet: “You want to tell me, and I have no objection to hearing it.”

By voice: Mrs. Bennet sits a little higher and forward, quick and breathless, every sentence leaning into the next. Mr. Bennet sits in the chest, slow and dry, with a small pause before each reply. The narrator stays neutral and lets Austen's own tag, “cried his wife impatiently”, do the rest.

With AI: cast a bright, eager middle-aged voice for Mrs. Bennet with [excited, rushing], and a low, mature voice for Mr. Bennet with [dry, unhurried]. Give the narrator an even, warm voice and no notes. Listen once, regenerate any line that misses, and the contrast is done.