Voice cloning

Clone your own voice from a short recording, then have it read text or voice-overs in other languages. You read a consent sentence first, and we check it before anything is cloned.

También en español

Practice the consent recording

To create your voice, you read a sentence like this one aloud. Practice here first: the recording stays in your browser and is not uploaded.

Example output

The speaker in this example is one of our own stock synthetic voices, so no real person's voice is cloned on this page. It read the English consent sentence below. That recording became the reference, and the clone then read a Spanish text it had never said.

Consent recording (English, 13.5 seconds)

I am recording my own voice and I agree to SoundHalo creating a synthetic copy of it. My check words are candle, garden and blanket. I will only use this voice for content I am allowed to make.

Cloned voice reading Spanish (6.6 seconds, 98 bytes, $0.0015)

Hola a todos. Gracias por venir hoy. Vamos a hablar del río pequeño que pasa detrás de mi casa.

Checks on this output: transcribing it gave back the Spanish text word for word, and its speaker similarity to the consent recording was 0.91 (the same-speaker threshold is 0.65; two different voices scored 0.42).

How the consent check works

  1. You confirm three statements: the voice is yours or you have the speaker's written permission, it is not a public figure, and the speaker is 18 or older.
  2. You get a sentence with three random check words. It expires after 15 minutes and works once, so an old or edited recording cannot be reused.
  3. You read it aloud in the browser. Our speech recognition must hear the sentence with at most a quarter of the words wrong. If it does not, nothing is cloned.
  4. Extra samples are compared with the consent voice by a speaker recognition model. Samples that sound like someone else are left out and you are told which.
  5. We store the consent recording, the sentence, what was heard and your confirmations with the date, as a record of permission.

This cannot prove who is speaking or that permission was really given. It makes casual misuse hard and leaves a record. Misuse breaks our terms and we will delete voices that break them.

Getting a good clone

  • Record in a quiet room, about a hand's width from the microphone, in your normal speaking voice.
  • No music, TV or other people in the background. Background sound gets copied into the voice.
  • Extra samples help most when they sound like the way you want the voice to speak: 5 to 60 seconds each, the same voice only.
  • The clone copies the sound of your voice, not your accent in every language. Listen to the whole file before you use it.

Limits and price

  • No account: practice the recording only. Cloning needs a free account.
  • Free account: $0.03 of cloned speech a day, up to 2,000 bytes per request (about 2,000 English characters, roughly two minutes of speech).
  • Pro ($15 a month): $0.30 a day, up to 20,000 bytes per request.
  • One personal voice per account. Up to three extra samples. The reference uses at most 45 seconds of audio.

The price is the voice provider's real cost per byte of text, with no markup on top of your plan. A request that reached the provider counts even if it fails. Limits reset at midnight UTC.

Languages

Your voice can read English (US), English (UK), Spanish, French, German, Italian, Portuguese (Brazil), Portuguese (Portugal), Japanese, Chinese (Mandarin), Russian, Turkish, Polish, Dutch, Vietnamese, Indonesian, Arabic. You record the consent sentence in English or Spanish; the clone can then speak any of these. These are the languages Fish Audio names for its model that our text to speech also covers. Other languages are not offered until we have checked them.

Use it in text to speech and voice-overs

Once your voice exists, choose "Your personal voice" in the voice-over tool when you are signed in, or type text on your voice page. The text to speech tool keeps its free stock voices. To dub a recording in the voices of the people in it, choose "Original speakers' voices" in the translate audio tool.

Acceptable use

Clone only your own voice or a voice you have written permission to use. Do not clone public figures, children, or anyone to deceive, harass, defraud or put words in their mouth. Do not use a cloned voice to get past voice checks at banks or other services. Everything made here is synthetic speech; label it where it could matter. See the terms.

Privacy

Your recordings are uploaded to SoundHalo. We keep one cleaned reference clip and the consent record in private storage until you delete the voice or your account; the uploads themselves are deleted once the voice is made. Each time the voice speaks, the reference, its words and your text go to Fish Audio, whose terms allow it to use submitted content to train its models. Speech recognition for the consent check uses our usual transcription provider. Generated audio follows our normal 30-day file access rule.

Related tools

FAQ

Can I clone someone else's voice? +

Only with their written permission, and never a public figure, celebrity, politician or anyone under 18. You confirm this before you start, and the person whose voice it is must read the consent sentence themselves. We keep that recording and your confirmation as a record.

How much does it cost? +

Cloned speech needs a free account. It is charged at what the voice provider charges us: $15 per million bytes of text, about $0.015 per 1,000 English characters. A free account includes $0.03 a day (about 2,000 English characters). Pro includes $0.30 a day. The cost of each request is shown before it runs.

Why does Japanese use more of my allowance? +

The provider bills bytes of text, not characters. English letters are one byte each. Japanese, Chinese and some other scripts use about three bytes per character, so the same number of characters costs about three times as much.

Which company processes my voice? +

Fish Audio. Each time your voice speaks, SoundHalo sends Fish Audio your reference recording, its words and the text to read. Fish Audio's terms allow it to use submitted content to improve its models, and they do not promise a deletion date. Do not create a voice if that is not acceptable to you.

Can I delete my voice? +

Yes, at any time from your account. The reference audio is deleted right away. The consent recording and record are kept for about 30 days as proof of permission, then deleted automatically. Deleting your account deletes the voice too. Copies already sent to Fish Audio follow its terms.

Is the audio marked as synthetic? +

The file carries a metadata comment saying it is synthetic speech in a cloned voice. Metadata is easy to remove, so tell your audience when they hear a cloned voice if it could matter to them, and follow any platform rules for labeling AI audio.

Can I use it to dub a video in my own voice? +

Yes. The voice-over tool can use your personal voice: give it subtitles or timed lines in any supported language. To dub a recording in the voices of the people speaking in it, use the translate audio tool and choose original speakers' voices.