AI voice cloning guide — clone your voice and make characters speak

A cloned voice is the difference between "an AI reads my script" and "I read my script, without booking a studio". It is also the fastest way to give a cast of characters voices that stay put across a hundred lines. This guide covers the practical side in the eroq Voice tool: what to record, how to upload it, which model to render with, and how to write text that the engine performs instead of recites.

First, whose voice

Yours, or one you have explicit permission to use. Cloning a real, identifiable person without consent — a celebrity, an ex, a streamer — is banned under the acceptable-use policy and enforced at the account level, which means the account goes, not just the clip. The platform is for adult fiction. Your narrator, your characters, your ad read: those are the jobs a clone is for.

Record a sample the engine can learn from

Voice cloning copies what it hears, including what you did not mean to give it. The tips below are honest about the engine rather than optimistic:

• A quiet room, no music, no reverb. Soft furnishings absorb echo; a closet full of coats is a better booth than a kitchen. Background hum becomes part of the voice.
• One mic, one distance. Keep your mouth at the same distance from the microphone for the whole take. Moving closer and further reads as two voices.
• Speak the way you want to be cloned. If the clone is for calm narration, record calm narration. If it is for a character who bites, give it a few lines with bite. The engine clones a manner, not just a timbre.
• Vary the sentences. Questions, statements, a list, an aside. Flat, repetitive text produces a flat clone.
• No processing. Skip the compressor and the noise gate; the raw file is the better input.
• Trim the dead air. Long silences at the start and end are wasted material.

Upload the file from the Voice tool. The clone shows up next to the roster voices — Aria (warm, intimate, female) and Orion (low, calm, male) — and can be used by both text-to-speech models.

Voice One or Voice Turbo

Two models render speech, and both use the same roster and the same clones:

• Voice One is the expressive flagship — 3 credits per 100 characters, MP3 out. Use it for anything with emotion in it: character lines, audiobook narration, a scene.
• Voice Turbo is faster — 2 credits per 100 characters. Use it for drafts, long neutral reads, UI copy, and iterating on a script before the final pass on Voice One.

Both are on the free tier, so the choice is about delivery and speed, not access. For a 1,000-character monologue, that is 30 credits on Voice One and 20 on Turbo — a difference small enough that the honest rule is "Turbo to iterate, One to ship".

Write for the ear — punctuation is direction

The voice engines read punctuation as delivery. There is no separate markup to learn; the script is the direction. A complete example, written to be performed:

You came back. I didn't think you would… not after last time. Sit down — no, really, sit. We have maybe ten minutes before they notice the door, and I am not spending them watching you stand there. So. Tell me what you saw.

What each mark does:

• Periods end a thought and drop the pitch. Short sentences read as decisive.
• Commas are short breaths. Too many and the delivery gets sing-song; cut them where you would not pause aloud.
• Ellipses trail off and slow down. One per paragraph, or the voice sounds unsure of everything.
• Dashes interrupt — a change of mind mid-sentence, a self-correction.
• Question marks lift the ending. Exclamation marks add energy; use them sparingly or everything shouts.
• One-word sentences ("So.") land as a beat. They are the closest thing to a director's pause.

Read the script aloud once before rendering. If you stumble, the engine will too, and 3 credits per 100 characters is cheap enough to fix the sentence and render again.

Speed and expressiveness

Two controls sit next to the model choice:

• Speed — slow down for audiobook narration and anything the listener needs to follow; speed up for ad reads and messaging-style lines.
• Expressiveness — low keeps the delivery even and close to the sample, right for documentary voice-over; high lets the engine lean into the punctuation, right for characters and drama.

Adjust one at a time. If a clone stops sounding like you, bring expressiveness down first, then speed; the sample is the anchor and both controls pull away from it.

Give the voice to a Character

A Character has a name, a persona, reference photos and a voice — and that voice can be your clone. Set it once and every line you generate for that character uses it; the Voice tool lists your cast, so you pick "Mara" rather than remembering which clone was hers. Pair it with the Text tool: write the scene with RP+, paste the line, render it in her voice, download the MP3 and lay it under the clip in your editor. TTS for AI characters goes deeper on writing lines that sound like a person rather than a paragraph.

Developers get the same voices over the API — audio/speech and voices (including cloning) are in the speech docs — and the eroq CLI renders a line in one command: eroq speech "Tell me what you saw." -v aria.

FAQ

Can I clone someone else's voice if they said yes?

Consent is the requirement, and it has to be real. Your own voice or a voice you have explicit permission to use is fine; a real person cloned without consent is an account-level ban.

Does the clone work on both voice models?

Yes. Voice One and Voice Turbo share the same roster and the same clones; the difference is expressiveness and price — 3 versus 2 credits per 100 characters.

How do I make the voice pause?

With punctuation. A period is a full stop, a comma a breath, an ellipsis a trail-off, a one-word sentence a beat. There is no separate pause tag.

Ready to hear yourself? Open the Voice tool — the first 50 credits are free, and both voice models are on the free tier.