Voice cloning explained: what it is and what it needs

Voice cloning means building a synthetic voice that sounds like one particular person. Wikipedia's definition of the closely related audio deepfake is technology designed to generate speech that convincingly mimics specific individuals.¹ For a business, the safe starting point is a voice that's your own, or one belonging to someone who has clearly agreed.

How much recording it takes

It varies by maker and by quality level. In 2018, researchers from Google presented work on speech that can be made to sound almost like anybody from a sample of only 5 seconds.² That was a research result. It's not a promise about any product.

Products tend to offer more than one level. ElevenLabs, as of its documentation checked on 2 November 2026, lists two.³

Plans change. The same page says a Creator plan or above is required for Professional Voice Clones, so check the maker's pricing page for what applies when you read this.³

What a responsible maker asks for first

Good makers don't just accept any upload. Two examples from their own documents:

• ElevenLabs says voice-captcha technology is used to verify that Professional Voice Clones are created from your own voice samples.³
• Microsoft's custom neural voice requires customers to upload a specific recorded statement spoken by the voice talent. In it, the talent acknowledges that their voice will be used to create synthetic voices. The statement's language must match the language of the training recordings.⁴

Why do these steps exist? A cloned voice can be misused, and these steps are how the maker records that the person agreed. Expect setup to involve more than pasting in a script. If a maker asks for a consent recording, make it in the language the maker specifies, and keep a copy.

Getting your own samples right

Makers publish their own recording tips, and you should follow them over anything here. The sensible general points are plain ones. Record somewhere quiet. Keep the microphone distance steady. Read naturally, in the style you want the clone to use. Deep neural networks are trained using a large amount of recorded speech, so what you record is what the model has to learn from.²

What you can't do with this page's information

Don't clone someone else's voice without their clear permission. That includes a former employee, a voice actor whose work you have on file, or a public figure. Permission, licence terms and the rules in your country are covered in consent and licence terms for synthetic voices.

Keep the recordings private. Microsoft's documentation warns that some jurisdictions impose special legal requirements on data such as biometric data, and that you must check the rules that apply to you.⁴ Ask a qualified adviser or your local regulator if you're unsure.