Can AI Make a Child's Voice?

Updated 2026-10-27·3 min read·Dialogue & audio

Why it's hard

Three things get in the way.

• Libraries are built from adults. Browse a voice library for "young" and you mostly get adult narrators who sound youthful. They come close, but a parent will hear the difference.
• Some providers refuse. At least one major voice provider won't produce child voices, so the option simply isn't there.
• Pitch-shifting an adult doesn't work. Raising an adult voice changes the pitch but not the size of the voice behind it, and the result is a chipmunk. We tried markup-based pitch controls too; they sounded worse.

What did work, when we tested options side by side:

• Start from a voice that's already young. A small shift of a real young voice, two or three semitones down, read convincingly as a younger boy. A small change to the right voice worked better than a big change to the wrong one.
• Design the voice from words. A voice generated from a description like "an eight-year-old girl with a Kenyan accent" came out as a believable child, accent included.
• Reuse the designed clip. Generating each new line from that same designed voice held the child's age and accent across lines. If you describe the voice from scratch each time, it drifts.

How to do it

1. Write the voice like a casting note. Age, sex, where they're from, how they talk: "eight-year-old girl, Kenyan English accent, bright and quick, a little breathless when she's excited."
2. Generate a short sample and listen. Judge it on a real line from your script, not "hello."
3. Lock it to the character. Once it's right, every line the character speaks should come from that one voice, across scenes and episodes.
4. Direct the performance separately. Keep emotion and delivery out of the voice description. Put them on the line itself, so the voice stays the same child whether she's laughing or crying.
5. Avoid pitching adults up. If you must shift pitch, shift a young source a little, never an adult a lot.
6. For stylized kids, say so. A cartoon sidekick can have a squeaky designed voice on purpose. That's a style choice and won't sound like a real child.

Example

A family drama needs a seven-year-old son, Sam, with four short lines across two scenes. Instead of hunting a library, write: "seven-year-old boy, American, soft-spoken, slightly hoarse, talks in quick bursts when nervous." Generate one sample on his first line, keep it, and voice all four lines from it. When Sam is scared in scene two, mark that line as frightened rather than rewriting his voice.

How ShowMaker handles it

In ShowMaker every character gets a fixed voice. Alongside ElevenLabs library voices, you can design a voice from a written description, including child voices and accents, and it stays that character's voice for every line. Delivery is set per line with tags (Express), so the performance changes and the voice doesn't. See AI character voices or the voices docs.