How to Clone a Voice Online

With Voice Clone you give the studio a short recording of a voice, type some text, and get an MP3 of that text spoken in the cloned voice. You use it in your browser and the audio is generated on our servers, so you don't need to install anything or have a graphics card. This guide covers each step, how to record a sample that works, and what to do when the result doesn't sound right.

What you need

• A Google account, which is how you sign in.
• A voice sample: an audio file (MP3, WAV, M4A, WebM or OGG, up to 10 MB), or a microphone to record straight from the browser.
• The right to use the voice: it has to be your own voice, or you need the speaker's permission.
• Credits: voice cloning costs 1 credit per 200 characters. New accounts get 20 free credits, valid for 365 days, which cover 4,000 characters of cloned speech in total, spread over several requests of up to 1,000 characters each.

How to clone a voice, step by step

1. Open the studio and sign in with Google.
2. Choose the Voice Clone tab.
3. Add your sample: press Upload audio and choose a file, or press Record, speak for 5–15 seconds and press Stop recording.
4. Tick the box confirming that this is your own voice or that you have the speaker's permission.
5. Type what the voice should say, up to 1,000 characters.
6. Choose the language your text is written in.
7. Check the credit cost shown in the studio, then press Generate.
8. Play the result in the page and download the MP3. It also stays in your recent history for 30 days, or 6 months once you've made a purchase.

To make more lines in the same voice, keep the sample and generate again. You can reuse the same sample for several generations in one session.

How to record a good voice sample

The sample decides most of what you hear, so a couple of minutes spent on it pays off.

• Length: the sample must contain at least 3 seconds of speech, and only the first 20 seconds are used. Aim for 5–15 seconds of continuous talking, and cut any long silence at the start so it doesn't take up part of those 20 seconds.
• One speaker: only the person you want to clone should be talking. Cut out other voices, laughter and interruptions.
• A quiet room: no music, TV, fan or traffic in the background. A room with curtains, a sofa or carpet has less echo than a kitchen or bathroom.
• Steady distance from the microphone: close enough that the voice is clear, but not so close that every breath pops.
• Speak the way you want it to sound: the clone picks up the tone and character of the sample. If you mumble or read in a flat voice, the result tends to sound the same. Talk at your normal pace, as if explaining something to a friend.

If you're not sure what to say, read something like this out loud:

I usually start the day with a cup of coffee and a short walk. Then I check the plan for the week, answer a few messages, and pick the one task that matters most.

Whose voice you can clone

You may only clone your own voice or a voice whose speaker has given you permission. The studio asks you to confirm this every time you clone. A recording being public, such as an interview, a podcast or a video online, does not count as permission. If you're cloning someone else's voice, it's sensible to have their agreement in writing, such as a message or an email. Laws on voice cloning also differ between countries and states.

Our Terms of Service don't allow using a cloned voice to:

• impersonate someone or commit fraud, including scam calls and fake voice messages;
• make deceptive or political deepfakes;
• harass anyone or create sexual content involving real people;
• pass generated audio off as a real recording of a real person.

Accounts that break these rules may be suspended.

Within these rules, audio you make with Voice Clone can be used commercially, for example in monetized YouTube or TikTok videos, ads, client work, paid courses and podcasts, and it carries no watermark.

The 21 languages Voice Clone speaks

Chinese, English, German, Spanish, French, Japanese, Italian, Hungarian, Korean, Russian, Persian, Arabic, Polish, Portuguese, Dutch, Czech, Danish, Swedish, Greek, Turkish and Hebrew.

Choose the language your text is written in from the dropdown. If your sample is in one language and your text in another, try a short line first, because results vary.

If your language isn't on the list, use an HD preset voice instead. HD voices read 29 languages, including Hindi, Indonesian, Filipino and Tamil.

Length limits and credits

• Text: up to 1,000 characters per request. For a longer script, split it at paragraph or sentence breaks into parts of up to 1,000 characters, generate each part with the same sample, and join the MP3 files in any audio or video editor.
• Sample: an upload can be up to 10 MB, and only the first 20 seconds of the sample are used.
• Cost: 1 credit per 200 characters, rounded up, with a minimum of 1 credit. A short line under 200 characters costs 1 credit, and a full 1,000-character request costs 5 credits.
• What credits buy: the 20 welcome credits cover 4,000 characters of cloned speech in total (each request still takes up to 1,000 characters), and 1,000 credits cover 200,000 characters. See pricing for plans and one-time packs.
• Failed jobs: if a generation fails, its credits are returned to your balance automatically.

If the clone doesn't sound right

Voice Clone runs on small open-source models: Pocket TTS by Kyutai for English, French, German, Spanish, Italian, Portuguese and Dutch, and MOSS-TTS-Nano, with 100 million parameters, for the other languages. They copy the tone and character of a voice from a short clip, but the result won't be identical to the original speaker. Within those limits, these fixes help most:

• It doesn't sound like the person. Record a new sample: 5–15 seconds of one person talking continuously, in a quiet room, with no music. The clone is made fresh from the sample on every request, so a better sample changes the very next result.
• It sounds muffled, noisy or echoey. Noise and echo in the sample carry over to the result. Record closer to the microphone and away from bare walls, or pick a cleaner part of your recording for the first 20 seconds.
• Words are mispronounced or the accent is off. Check that the language dropdown matches your text. Write numbers, dates, abbreviations and symbols out as words, for example "twenty twenty-six" instead of a year in digits.
• It sounds flat or rushed. Record the sample with the energy and pace you want to hear. In your text, commas and full stops give the voice places to pause, and very long sentences are easier to read when split.
• One take is good, another isn't. Generate the line again and keep the version you like.

Still stuck? Email [email protected] with the tool you used and the text you typed.

Voice Clone or a preset voice?

Voice Clone and Standard preset voices cost the same: 1 credit per 200 characters.

Use Voice Clone when you need a particular voice, such as your own voice for narration that sounds like you, or when your language is one of the clone languages that Standard doesn't speak, like Japanese, Korean, Chinese or Arabic.

Use a Standard voice when any clear voice will do. There's no sample to record and no consent step, each request takes up to 3,000 characters instead of 1,000, and all 23 voices speak English, French, German, Spanish, Italian, Portuguese and Dutch. You can play a preview of every voice in the voices section before you generate.

Use an HD voice when you need one of its 29 languages, speed control from 0.7× to 1.2×, or up to 5,000 characters per request. HD costs 3 credits per 100 characters, and the text you type for it is sent to a third-party voice provider.

What happens to your sample

Voice Clone runs entirely on our own servers. Your sample is not sent to any third-party AI company, and we don't train a custom model on it or use it to train models. The sample and the audio you generate are kept in our cloud storage for your account for 30 days so your links and history work, then deleted automatically; after you've made a purchase, new samples and results are kept for 6 months. Download any audio you want to keep, and email [email protected] if you'd like them deleted sooner. The Privacy Policy has the details.

Ready to try it? Go to the studio and open the Voice Clone tab.