AI character voices that stay the same person

What a character voice has to do

Most text-to-speech tools treat a voice as a setting you pick for each clip. That works for a single narration. It breaks the moment you have five characters in a scene and forty scenes in a season, because nothing ties a voice to a person. Pick a slightly different voice in episode six and the audience hears a recast.

In ShowMaker Studio the voice lives on the character record, next to their character sheet and wardrobe. Every line that character speaks, in any scene or episode of a series, is voiced with it.

That matters most in a series, where the cast is shared across every episode. A side character who turns up again in episode nine sounds the way they did in episode two, without anyone having to remember which voice was used. It also means a recast is a single change: swap the voice on the character and every line they speak is voiced again with the new one.

Library voices or designed voices

For most adult roles, a library voice is the fast answer. You audition a few, pick the one that fits, and you're done. The harder roles are the ones libraries don't cover well: children, specific regional accents, an elderly voice with a particular texture.

For those, ShowMaker can design a voice from a written description. You describe who is speaking, how old they are, where they're from and how they sound, and a voice is made to match. That voice is then kept for the character like any other.

Write the description the way a casting director would brief an actor: age, where they grew up, the energy they bring into a room, anything distinctive about how they speak. A vague brief like a friendly woman gives you a generic voice. A specific one, like a tired nurse in her fifties from Lagos who speaks slowly and warmly, gives the model something to aim at.

• Library voices from ElevenLabs for fast casting
• Designed voices from a plain description
• Child voices and accents, which libraries rarely offer
• One fixed voice per character, shared across a series

Direction is what makes a voice sound acted

A correct voice reading a line neutrally still sounds like a machine. Drama needs delivery: the line a character says through clenched teeth, the one they barely get out. Express lets you add delivery tags to a line so the performance carries what the scene means as well as the words.

A good rule is to write the emotion into the line's direction as well as the text. Short lines with a clear delivery tag tend to perform better than long lines asked to carry several shifts at once.

Think about who is listening, too. A line shouted across a room and the same line whispered to someone a foot away are different performances, and the shot they sit in should agree. If the scene changes mood halfway through a speech, split the speech into two lines and direct each one separately.

What usually goes wrong

Binding the voice to the character stops drift, and direction fixes a flat read. Timing is the problem people miss. A line that runs longer than its shot gets cut off or rushed. ShowMaker's checks flag a line that is too long for the shot it sits in, so you can split it or lengthen the shot before you pay for video.

Voices are also the cheapest part of the film to fix, so settle them early. Listen through the animatic, recast anyone who doesn't fit, then render.

One more check is worth doing with the whole cast together: listen to a scene with every speaker in it. Two voices that sounded right on their own can sit too close together in pitch and pace, and the audience loses track of who is talking. If that happens, change one of them now, while it's one setting on one character.