Voice Cloning Tutorial: One Minute of Audio, Your Voice

Ninety percent of a clone's quality comes from the sample, not the button. This walkthrough follows the order you actually work in: what to record, how long, how to check the result, and when you should design a voice instead of cloning one.

Step one: record the sample

• Length: one to three minutes is enough; past five the gains are tiny.
• Content: read something natural in the way you normally speak, not a news bulletin.
• Room: quiet, phone about 20 cm from your mouth, air conditioning and fans off.
• Avoid: background music, other people, echoey rooms.

Step two: upload and name it

Upload on the voice cloning page and give the voice a name you will recognise. Cloning usually finishes within a minute.

Step three: check it

Test three kinds of text: a plain statement, an emotional line of dialogue, and a sentence with numbers and brand names. Listen for three things: does it sound like you, is there any hiss, does it drop words in long sentences.

Step four: tune

• Stability too high sounds robotic, too low gets erratic; start at 0.5.
• Similarity above 0.8 sounds more like you but also amplifies flaws in the sample.
• Emotion comes from tags, not sliders — see the emotion tags cheatsheet.

When not to clone

If what you need is "a professional voice" rather than "my voice", design one: a sentence of description produces a person who does not exist, with no consent questions attached.

Voice: AliceEnglish5 s

Voice: 英伦绅士管家Chinese9 s

The consent line

Clone only your own voice or one you have written permission for. Celebrities, colleagues and family recordings are all off limits; details in the voice cloning consent guide.

A finished clone works directly in text to speech, audiobooks and video dubbing.