AI Character Voices: Design a Voice From a Description

VdoBloom’s Voice Design tool creates an AI character voice from a written description rather than from a preset list. You describe the voice — age, accent, tone, pace — the tool generates it, and then you type any script and hear it read back in that voice as an MP3. Designing a voice costs nothing separate from the read: speaking 1,000 characters costs 2 credits and the 2,000-character maximum costs 3 credits, and you get 10 free credits at signup with no card.

Updated October 2026.

What the Voice Design tool actually does

The tool has two steps on one page, and they are worth separating in your head before you start. Step one is design: you write a description of the voice you want and the model builds it. Step two is speak: you pick one of the voices you have designed, paste text, set a speed, and generate audio.

The design step takes three inputs. The description is required, up to 1,000 characters, and it is where all the character lives. A sample line is optional, up to 300 characters, and it is read back to you so you can hear the voice immediately. A name is also optional, up to 60 characters, and it is only so you can find the voice again in your list later.

The speak step takes your chosen voice, text up to 2,000 characters, and a speed slider from 0.5 to 2.0 in steps of 0.05. The output is an MP3.

The language selector covers Auto, English, Chinese, Japanese, Korean, German, French, Russian, Portuguese, Spanish and Italian. That is 11 options, and it is worth naming them because language coverage is exactly where character-voice tools tend to over-promise.

Step 1: describe the voice

1. Open the Voice Design tool and look at the first box, labelled “1. Design a voice”.
2. Write a description of at least five characters. The field takes up to 1,000 characters. You do not need all of them, but a single adjective gives you a generic voice, not a character.
3. Add a sample line if you want to hear it straight away. This is optional and capped at 300 characters. Write something in the voice’s own rhythm rather than a neutral sentence — the sample is your first listen.
4. Name the voice. Up to 60 characters. “Gruff dockmaster” is more useful to you next month than “voice 3”.
5. Pick the language. Choose from the 11 options, or leave it on Auto if the script is English.
6. Generate the voice. Once it is created it stays in your list of designed voices.

Step 2: speak with the designed voice

1. Pick your voice from the dropdown of everything you have designed.
2. Paste the script into the text box, up to 2,000 characters. The counter under the box tells you where you are.
3. Set the speed between 0.5 and 2.0. Slower is not automatically more menacing and faster is not automatically more energetic; match the character, then stop fiddling.
4. Check the credit cost, which the tool shows before you generate.
5. Generate and download the MP3.

Voice description examples

These are written the way the field expects: physical detail, delivery, and one behavioural note that gives the voice something to do.

• “A warm middle-aged woman with a soft Irish accent, unhurried, speaking as if explaining something to a friend across a kitchen table.”
• “An old male narrator, gravelly and low, with long pauses between sentences and a habit of trailing off at the end of a thought.”
• “A precise young man in his twenties with a clipped British accent, fast and clinical, like a systems engineer reading a status report.”
• “A cheerful teenage girl with a light American accent, high energy, sentences that tumble into each other when she gets excited.”
• “A calm female newsreader, neutral accent, even pace, no emotional colour at all.”
• “A deep male villain voice, slow and deliberate, savouring each word, with a faint Eastern European accent.”
• “An exhausted night-shift dispatcher, middle-aged man, flat Australian accent, speaking quietly and slightly hoarse.”
• “A bright children’s storyteller, female, expressive and unhurried, lifting at the end of each sentence to invite a response.”

Why describe a voice instead of picking a preset

Preset voice libraries have a shelf life. After a while the same dozen voices turn up in every podcast, tutorial and YouTube short, and an audience starts to hear the tool rather than the character. A description-driven voice gives you something that is not on a menu — and, more practically, it gives you a voice you can rebuild from the same words if you need it again in six months.

It also solves a specific problem in series work. If you are making an audio drama, a game demo or a short-form series, you need the character to sound the same every episode. A written description is a specification you can keep, edit and hand to someone else. “The voice from last time” is not.

Common mistakes

• Describing the emotion instead of the voice. “Angry” is a line reading, not a character. Describe the voice first, then write lines that are angry.
• Stacking contradictions. “A soft, booming whisper with a bright, gravelly warmth” gives the model nothing to hold on to. Two or three clear attributes beat six clever ones.
• Forgetting accent. Accent is the single fastest way to place a character, and it is often the detail people leave out.
• Writing the sample line as a test sentence. Use the sample to test the character, not the microphone. A neutral sentence tells you almost nothing.
• Fixing pacing with the speed slider. If a line feels wrong at 1.0, the words or the description are the problem. Speed is for matching a runtime, not for rescue.
• Reusing one voice for every character in a scene. Two characters in a conversation need two distinct voices, or listeners lose track of who is speaking within about four lines.

Credits, limits and what this tool is not

Designing a voice is free; you pay for the speech you generate. The rate is 0.0012 credits per character with a one-credit minimum, so a 1,000-character read costs 2 credits and the full 2,000-character box costs 3 credits. A short 300-character line costs 1 credit. Because the cost is shown before you generate, you can check any script before committing.

The hard limits are the 1,000-character description, the 300-character sample line, the 60-character name and the 2,000-character script, with speed from 0.5 to 2.0. Output is MP3.

One thing to be clear about: this is a voice creation tool, not a voice cloning tool. VdoBloom does not clone real people’s voices in the audio tools — voices are built from your description only. That is a real limitation if your goal is to reproduce a specific person, and it is also the reason the tool is safe to use on voices that do not exist. If you want a library of ready-made narrators instead, the AI voice generator offers four engines and dozens of stock voices, which is a different job.

Frequently asked questions

Is there a free AI character voice generator?

Yes, up to a point. VdoBloom gives every new account 10 free credits with no card required, and designing a voice costs nothing — you only spend credits when you generate speech. A 300-character line costs 1 credit, so 10 free credits is enough to test several characters before you pay anything.

How much text can one AI character voice read?

A single generation on VdoBloom caps at 2,000 characters. For longer scripts, split the text at paragraph breaks and generate in sections with the same designed voice, then join the files. The voice itself can be reused for as many generations as you like.

Can I clone a real voice with this tool?

No. VdoBloom’s Voice Design builds a voice from a written description, and cloning real people’s voices is not supported in the audio tools. If you need a specific existing voice, this is the wrong tool rather than a tool with a setting to change.

What languages can designed voices speak?

Eleven: Auto, English, Chinese, Japanese, Korean, German, French, Russian, Portuguese, Spanish and Italian. Auto works well for English scripts; picking the language explicitly is safer when the script contains names or terms that should be pronounced a particular way.

How many characters should I make for a story or game?

Enough that listeners can tell them apart with their eyes closed. Two voices fail when they sit in the same register, so spread them across age and pitch as well as accent. A practical rule for a two-character scene is one voice that is higher and faster and one that is lower and slower — the contrast does more work than any amount of extra description.

Try it on a single line first: open Voice Design, describe one character, and let the sample read tell you whether the description is doing its job. If you would rather start from stock voices, AI text to speech explains how those libraries are organised, and when the voice needs to sit on a face, read AI voice and lip sync. Compare plan sizes on pricing before you scale up a series, and for where synthetic voices are heading, our AI voice overview is a good next read.