Change a voice with a neural network

Changing a voice means altering its timbre, gender, age or "character" while keeping the words and intonation intact. It's the basis of anonymity, characters for voice-overs, entertainment content and streams. Unlike cloning, you don't need a sample of a specific person — you simply turn one voice into another.

How it works

The network splits a recording into two parts: what was said (the words and intonation) and who said it (the timbre, the character of the voice). It keeps the first and replaces the second with the voice you chose. So your speech, emotion and rhythm stay, but a different person is heard. This is called voice conversion.

Upload a recording (or speak one in) — pick a new voice and listen to the result. Your first generations are free after signing up.

What you can change

• Gender — male ↔ female.
• Age and character — younger/older, softer/harder.
• Character — a robot voice, a fairy-tale hero, a narrator.
• Anonymization — hide your voice while keeping the speech (for an interview, say).

To get a clean result

• A clean recording with no noise. Background and echo carry into the result — record in a quiet place.
• Natural speech in the original. Conversion keeps the intonation: speak expressively and the result will be alive.
• A target voice close in range. Turning a very low voice into a very high one sounds less natural than a moderate shift.

Changing a voice vs cloning

A common mix-up. Changing a voice turns speech into some other voice (a ready preset or an abstract one). Cloning a voice reproduces a specific real person's voice from a sample. The first is about style and anonymity; the second is about an exact copy (and far more sensitive ethically).

Important: ethics

Changing your own voice or using abstract presets is harmless. But turning your speech into the voice of a specific real person without their consent is already the territory of cloning and its risks (more on that in the next chapter). Don't use voice changing to deceive by impersonating someone else.

What's next

If you need not "some other voice" but an exact copy of a specific one, that's cloning — the most powerful and most sensitive feature in this guide.

In the Twelver chat you can upload a recording and change the voice right in the conversation. A few generations are free after signing up.