Clone Your Voice: AI Voice from a Short Recording

Record yourself, choose a clip, or pick part of a longer recording. Type any text and download it spoken in your voice. Your recording never leaves this device.

  1. 1Choose the language
  2. 2Add your voice
  3. 3Type the text and get the audio

Choose it first: you record in this language, and your voice reads text in this language only. Changing it later clears your voice.

Voice model (English) · 158.1 MBDownloads once, then stays in this browser.

Add your voice first.

Add your voice · English

12 to 25 s of natural speech in this language is enough. Your recording stays on this device.

Watch it in action

Clone your voice from a short recording and hear it read any text, right in your browser.

How to clone your voice

1

Choose the language, then add your voice

Pick the language you will speak. Then record yourself, choose a clip, or pick part of a longer recording.

2

Check the consent box

Confirm that the voice is yours, or that you have the speaker's explicit consent. Your voice is not prepared until you check the box.

3

Type the text

Type or paste what your voice should say, in the language you recorded in.

4

Download the AI voice audio

Listen to the result, then download it as MP3, M4A, or WAV. The file is named speech-ai-voice, so it is never mistaken for a real recording.

Clone Your Voice: AI Voice from a Short Recording

SubtitleKit's voice clone tool turns a short recording of your voice into an AI voice that reads any text you type, then hands you the audio file. Record yourself in the browser, choose a clip, or pick part of a longer recording, confirm that the voice is yours or that you have the speaker's consent, type the text, and download it spoken in your voice as MP3, M4A, or WAV. Everything runs inside your browser, so it is free, needs no account, and your recording never leaves this device.

How voice cloning works here

The tool first finds the speech in your recording and trims the silence and anything past the length the model needs. An AI model running on your own machine then turns that speech into a voice embedding, a compact description of how your voice sounds. The same model reads your text line by line in that voice, and the lines are joined into one audio file in the browser tab. The voice embedding is kept in this browser so the next run skips that step; the recording itself is never stored.

Free, no sign-up, no watermark. The only wait is the first run, when a 158 MB AI voice model downloads once and is cached in your browser. Every later run starts from the cache, and generation itself works with no internet connection.

What is voice cloning?

Voice cloning builds a synthetic voice that sounds like a real person from a recording of them speaking. Most online voice cloners ask for an account, send your recording to their servers, and charge by the minute once a free trial runs out. This one runs the model in your browser instead, so your voice recording and your text stay on this device and there is no meter running.

Which languages your voice can speak

Your voice can speak English, Spanish, French, German, Italian, and Dutch. Each language has its own model, so a voice speaks the language it was recorded in; to hear your voice in Spanish, record yourself speaking Spanish. Portuguese has a model, but it does not sound enough like the speaker yet, so it stays switched off until it does. Other languages are listed in the picker with the reason they are not available yet.

Your voice, with consent

Before your voice is prepared, you confirm: "This is my voice, or I have the speaker's explicit consent. I won't present the output as a genuine recording." The result and the downloaded file are labeled as an AI voice. Cloning someone without their consent, or passing the audio off as a real recording of a person, is not allowed by the model's license or by this page.

What to expect

The first run downloads a 158 MB voice model, which takes a few minutes on a slow connection. Preparing your voice the first time takes a while on an older laptop, and a desktop browser is the most reliable place to do it. Text dense with numbers, dates, or symbols reads poorly, so spell them out the way they should sound. A voice speaks only the language it was recorded in; reading English text in a voice recorded in French is not available yet.

Who uses this tool

Content creators

Turn a script into a narration in your own voice without a recording session, or fix one line of a voiceover without recording the whole take again.

Accessibility

People who are losing their voice, or who find speaking tiring, can keep a voice that sounds like them for messages and recordings.

Teachers and learners

Produce course audio in your own voice from written notes, and keep it consistent across lessons recorded weeks apart.

Writers and editors

Hear a draft read back in your own voice to catch awkward sentences, or make a quick scratch narration before the final recording.

Frequently Asked Questions

Can I clone someone else's voice?

Only with their explicit consent. Before your voice is prepared, you check a box that reads: "This is my voice, or I have the speaker's explicit consent. I won't present the output as a genuine recording." Your voice is not prepared until you check the box.

Which languages does it support?

English, Spanish, French, German, Italian, and Dutch. Portuguese has a model, but it does not sound enough like the speaker yet, so it is switched off until it does. Other languages are listed with the reason they are not available yet.

Does my recording leave my device?

No. The recording and the text never leave the browser. Only the voice embedding, a compact description of how the voice sounds, is kept, and only in this browser, until you tell the tool to forget it or clear this site's data in your browser. The recording itself is never stored.

How long should my recording be?

Record in a quiet room, about 12 to 25 seconds of natural speech in English, or 10 to 25 seconds in the other languages. You can also choose a clip you already have, or pick part of a longer recording, such as a podcast or an interview: the tool suggests the part with the most speech, and you can move it before you use it. Extra is trimmed, and the meter on the page shows when you have enough for the language you picked.

Is the audio marked as AI?

Yes. The result is shown with an AI voice badge, and the downloaded file is named speech-ai-voice. You must not pass the audio off as a real recording of the person whose voice it is.

Which model does it use, and under what license?

The voice model is Pocket TTS by Kyutai, and its weights are licensed under CC BY 4.0. Kyutai prohibits cloning a voice without explicit and lawful consent, presenting generated audio as a genuine recording of real people or events, and any illegal, deceptive, harassing, or privacy-invasive use. The full license is linked from the card that asks before the one-time model download.

What does it not do well yet?

Text full of numbers reads poorly, so spell numbers out as words. A voice also speaks only the language it was recorded in, so it cannot yet read English text in a voice recorded in French.

What does it cost, and what do I need?

It is free and needs no account. The first run downloads a 158 MB voice model once, which is cached in your browser after that. A desktop browser is recommended, because preparing a voice needs more memory than most phones give a single tab.

Tips for the best results

Record in a quiet room. Speak naturally, at your normal pace, with no music or other voices in the background. A clean recording clones better than a long one.

Spell out numbers, dates, and symbols. Write "twenty twenty-six" rather than "2026" if that is how it should sound.

Record in the language you want to hear. Your voice speaks the language of the recording, so record in Spanish to generate Spanish.