Voice cloning, done legally and done well

What a clone is - and is not

A clone captures who is speaking: the timbre, the resonance, the general placement of the voice. It does not capture how they speak - the timing, the emphasis and the habits that make someone recognisable to their own family. That is why a clone of your own voice reading a new script sounds like you and does not sound like you talking.

If what you actually want is your performance in a different voice, use the voice changer instead: you act the line, the model swaps the timbre, and your timing survives.

What a usable reference sounds like

• 30-60 seconds of continuous speech. More is not better; cleaner is better.
• One speaker, no music, no laughter, no phone.
• Normal delivery. Reference a shout and everything you generate will shout.
• Same language as the scripts you intend to read, where you can.
• A quiet room and a fixed distance from the mic. Room echo is copied faithfully, and it is the most common reason a clone sounds thin.

Consent is the part that has teeth

Cloning a voice that is not yours without documented permission is how services get sued and clients get dropped, and in a growing number of jurisdictions it is unlawful outright. Naadly makes the paperwork part of the flow rather than a checkbox: before a clone is created it asks for the speaker's name, the relationship, and a recording of the speaker reading a consent statement aloud. The record keeps a hash of the reference clip - proof that the consent attaches to this recording - and it survives account deletion, because it exists to protect the person who was cloned rather than the person who did the cloning.

A clone can be deleted at any time; the consent record remains, and either party can write to us to ask what is held.

When cloning is the wrong answer

• You need emotion - use the voice changer with an acted reference.
• You need one consistent brand voice forever - a catalogue voice never leaves the company, changes its mind, or raises its rate.
• You need forty voices for a video game - cast from the catalogue.
• You want a celebrity or a colleague who has not agreed - no.

Step by step

1. Record 30 clean seconds One speaker, quiet room, normal delivery, no music. WAV or MP3.
2. Record the consent statement The speaker reads the statement aloud in the same session. Naadly stores it with a hash of the reference clip.
3. Create the voice Upload both, name the voice, and it appears in your own catalogue alongside the public voices.
4. Test it on a hard line Numbers, a name, a question. If it sounds thin, the reference room was the problem, not the model.
5. Use or delete Render with it like any other voice, or delete it; the consent record stays as the evidence it was allowed.

Questions

How much audio do I need to clone a voice?

About 30 seconds of clean single-speaker speech. Quality of the recording matters far more than length.

Is voice cloning legal?

Cloning your own voice, or someone's with documented consent, is legal in most places. Cloning an identifiable person without permission exposes you to personality, publicity and - increasingly - specific AI statutes. Naadly requires a spoken consent record.

Can I clone a voice for free?

Not on Naadly: cloning is on paid plans from $15 a month. Free cloning tools generally pay for themselves with your recording.

Do you train models on my voice?

No. Reference audio is used to create your clone and nothing else, and it is not used to train any model.