A character voice for narration

An audiobook, a video with several characters, a course where one voice asks and another answers. The job sounds like “make a character’s voice”, and it has three solutions at different prices. The most obvious one is not the one that serves you best. Let us go through them in order.

Three routes and what they actually give you

Route	What it needs from you	What to expect
A ready-made voice from the set	Nothing: pick one of more than forty	Instant, with no preparation. The voices are professional and varied in character, but they are narrators, not characters
Describe the voice in words	A written description of the timbre, in English	A new voice made to your description and saved, so every later line sounds like the same speaker. Good for an archetype: storyteller, villain, robot. Takes about half a minute
Record and clone	20 seconds of you speaking as the character	The cheapest and the most controllable. And the only route where you decide how the character sounds instead of guessing at a phrasing

A description suits an archetype; a recording suits a character who must sound exactly one way.

On “describe the voice in words”, honestly

Describe the voice in one or two sentences: gender, age, timbre, pace. The model creates a new voice from it and saves it, so the next chapter, episode or line is read by the same speaker. Creating it costs about two to four times as much as a clone, paid once; after that the voice costs only the narration.

A description gets you a voice in the spirit of your words, not the one you hear in your head. Child voices are not created from a description — those, like any voice that must sound exactly one way, are more reliably acted by you and cloned.

Why “act it and record it” is the best option, not a fallback

Price aside, there is a stronger argument. You know how your character sounds, and a description tries to pass that knowledge through the narrow neck of a few adjectives. Twenty seconds of you speaking as them hold everything at once: pace, pitch, manner, the typical pauses, the very intonation you could not have described.

You do not need to act professionally. It is enough to speak in character: slower and lower for an old storyteller, faster and higher for a child, clipped for a soldier. The model captures the manner, not the acting skill.

Several characters, several recordings

For a dialogue, record each character separately, twenty seconds each. The voices are saved independently, and after that each line is read by the right one, even alternating line by line.

Record everyone in the same room on the same microphone: if one character was recorded in a kitchen and another in a car, the dialogue will reveal it more than the difference between the voices.

A recording that carries a character

Is your recording good enough

A clone comes out exactly like the twenty seconds it was made from. Tick everything that is true of your recording and see what you will hear later.

This recording will do

Upload it. The voice will be saved and stay available for any text, so you will not need to record again.

And one rule that has nothing to do with recording quality: you may copy your own voice, or someone else’s with their consent. A recording taken from someone else’s video is still their voice, however easily it uploads.

On emotion in the text

One more thing worth knowing in advance. The engine has a separate delivery field, “read it excitedly”, “slow and grim”, and it works for English text, but only with some of the ready-made voices. A cloned or designed character voice does not take it: that voice already carries its own manner. So for a character, emotion comes from two other places, and both are in your hands:

From the recording. The clone inherits the manner it was recorded in; that is the main lever.
From the text itself. Short sentences speed the speech up. A full stop instead of a comma gives a longer pause. An ellipsis sounds thoughtful. The model breathes by punctuation, so punctuation works here like a stage direction.

Poll

What do you need a character voice for?

Vote to see the results

Send twenty seconds in character, and hear what comes out.

Loading…

What to read next

How the whole job works: clone a voice and use it to read your text. What makes narration sound alive: reading a text aloud with a voice. If the character has to talk for a long time: long-text narration and its limits. And if the character also has to appear on screen, that is a neighbouring book: a talking photo.

Frequently asked questions

How do I have a text read in a character’s voice?

The most reliable way is to record 10–20 seconds of yourself speaking as the character and clone that recording: the clone inherits the pace, pitch and manner. You do not need to act professionally; speaking in character is enough. The voice is then saved and reads any lines you give it. You can try it in the Twelver chat.

Can I just describe the voice in words?

Yes. Describe the voice in a sentence or two and the model creates a new voice and saves it, so every later line is read by the same speaker. Creating it costs about two to four times as much as cloning a recording, paid once — enough for an archetype like a storyteller or a robot. A child voice, or one that must sound exactly one way, is more reliably acted by you and cloned.

How many different voices can I make?

As many as you like: each recording gives a separate saved voice, and they do not interfere with each other. For a dialogue, record each character for twenty seconds and have each line read in the right voice. Record everyone in the same room on the same microphone, because a difference in acoustics is more audible in a dialogue than a difference in voices.

Are there ready-made voices if I do not want to record?

Yes, more than forty, varied in gender and character, and every one of them speaks English. They are professional narrator voices, an excellent choice for a how-to, reference material or a voice-over. For a character with a recognisable personality, your own recording usually lands closer to what you had in mind.

How do I set an emotion: fear, joy, tiredness?

For a cloned character, in two ways, and both are in your hands. First, record the sample in the manner you want; the clone inherits it. Second, the text itself: short sentences speed speech up, a full stop instead of a comma gives a longer pause, an ellipsis sounds thoughtful. The engine does have a separate delivery field for English text, but it applies only to some of the ready-made voices, never to a cloned or designed one.

Can I make the voice of a famous actor or a cartoon character?

No, and that is not a technical limit but a line this book does not cross. A specific person’s voice belongs to them: you may copy your own voice, or someone else’s with their consent. For a fictional character, the right route is to play them yourself; it is both lawful and, as a rule, closer to what you had in mind.

In the Twelver chat you can record a character in twenty seconds and then have them read any lines, even a dialogue between several characters. Starter tokens arrive on your first visit, and signing up adds more.

Try it yourself

Everything in this guide runs inside Twelver

One chat for text, images, video, music and voice — no separate services or subscriptions.

Open Twelver chat

Related pageVoice Cloning

Rate your experience