AI voice cloning for text to speech: clone your own voice
VoiceLabs clones a voice from a recording of clear speech, 2 to 30 seconds long, and then speaks any text you type in it. The cloning and the speech run on our GPU servers, not in your browser, so there is nothing to install. The voice stays on your account and works in the studio, through the API, in Claude and ChatGPT, and in the browser extension.
How to clone a voice
In the studio, the Create Voice dialog takes four things.
- A recording. Upload an audio file, or record in the browser. It must be between 2 and 30 seconds of clear speech.
- The words spoken in it. Type them, or press Transcribe and let the studio fill them in. The engine aligns the voice against this text, so it should match the audio. It may be up to 1,000 characters.
- A name and a language. The language is the one the recording is in. You can also choose the engine that should voice it by default.
- Text to speak. Save the voice, pick it, type or paste any text and generate. The voice is on your account, so the API, the MCP server and the browser extension list it as well.
How long the recording has to be
2 to 30 seconds, and the engine enforces both ends: a shorter or a longer recording is refused with a message that names the limit. The length is measured after the silence at the start and the end is trimmed.
What a recording is refused for
- Shorter than 2 or longer than 30 seconds.
- Silent, or too quiet to hear.
- Not an audio file the engine can decode.
- Through the API, larger than 10 MiB. It takes WAV, MP3, M4A, OGG, FLAC, AAC, WebM and Opus.
Which engines clone a voice
Of the seven engines, five clone a voice from a recording. Qwen3-TTS CustomVoice and Kokoro speak built-in preset voices and do not take one. For a voice you save from a recording, the studio offers only the engines that can clone it.
- Qwen3-TTS10 languages
- LuxTTSEnglish only
- Chatterbox Multilingual23 languages
- Chatterbox TurboEnglish only
- TADA10 languages
The engines page lists the open-source project behind each one, and the language index says which engines speak which language.
Languages, Hindi included
VoiceLabs speaks 23 languages in all, and Chatterbox Multilingual is the cloning engine that speaks every one of them: Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, Italian, Hebrew, Arabic, Danish, Greek, Finnish, Hindi, Malay, Dutch, Norwegian, Polish, Swedish, Swahili, Turkish.
Hindi and Marathi
- Hindi is supported. Chatterbox Multilingual speaks it and clones a voice from a recording. Kokoro also speaks Hindi, but only in its built-in preset voices, so it cannot clone yours. No other engine speaks Hindi. The Hindi page has the detail.
- Marathi is not. It is not one of the 23 languages VoiceLabs accepts, so no engine here is set up for it, cloned or not.
The other cloning engines speak fewer languages: see the list above. English is the only language that all five cloning engines speak.
In the cloud, not in your browser
Your recording is sent to VoiceLabs, the models run on our GPU servers, and the audio comes back to your browser tab or your code. Nothing is computed on your device, so there is no model to download and no graphics card is needed. The trade-off is that your recording and your text leave your machine; if they must not, a model you run yourself is the right choice. The cloud text to speech page sets the two side by side, and the privacy policy says how a recording is handled.
That is also how your voice reaches a web page. A browser cannot add a voice of its own to its built-in read-aloud, so the read-aloud extension for Edge and Chrome asks VoiceLabs to make the speech and plays the result as audio. To play it in a page of your own, see browser text to speech.
Use your cloned voice anywhere
In the studio
The VoiceLabs studio is a browser tab: clone a voice, write, choose an engine and download what you make.
Through the API
POST /v1/voices/clone creates a voice from a recording, with a key that carries the voice:clone scope. The @voicelabs/sdk package wraps it as cloneVoice. The API page has the requests.
In Claude, ChatGPT and Cursor
The MCP server speaks in your voices, and its clone_voice_profile tool creates one from a recording once you approve the voice:clone permission. The MCP server page has the setup.
Reading the web aloud
The extension for Edge and Chrome reads a page or a selection aloud in a voice from your account, a clone of your own included.
Consent and responsible use
VoiceLabs cannot check whose voice a recording is, so the rule is yours to keep. The responsible use policy requires consent: by cloning or generating a voice you affirm that it is your own voice, or that you have the explicit permission of the speaker, or that the material is otherwise legally authorized.
- The studio asks first. Before you first use it, it shows “Only clone your own voice, or a voice you have explicit permission to use” and records that you agreed.
- The API and the MCP server have their own permission. Cloning needs the
voice:clonescope, separate from generating speech. In an AI assistant the consent screen asks “Create a cloned voice from a recording you provide”. The API reference adds: only clone a voice whose owner has agreed to it. - What is not allowed. Impersonating someone without permission, fraud, scams and phishing, harassment, and misleading political, legal, financial, medical or emergency communications.
Audio rights says what you may do with what you generate, and you can report misuse.
What it costs
One plan: VoiceLabs Pro, US$8 a month or US$59 a year, with a 7-day trial that asks for a card up front. It includes voice cloning with unlimited saved cloned voices, all seven engines, API access and 2 hours of generated audio a month (fair use). It is a flat subscription with no per-use fee. The pricing page has the detail.
Questions about voice cloning
- How does voice cloning for text to speech work?
- You give VoiceLabs a recording of clear speech, 2 to 30 seconds long, and the words spoken in it. It creates a voice on your account, and from then on you pick that voice for any text you type: the speech comes back in a voice built from your recording. The cloning and the speech both run on our GPU servers, so there is nothing to install. The studio does it in a browser tab, and the API and the MCP server can do the same. Clone only a voice you have the right to use.Cloud text to speechResponsible use
- How long a recording do I need to clone a voice?
- Between 2 to 30 seconds of clear speech. A shorter or a longer recording is refused with a message that names the limit, and so is one that is silent or too quiet. The length is measured after the silence at the start and the end is trimmed. You also give the exact words spoken, up to 1,000 characters, and the engine aligns the voice against that transcript, so it should match the audio. Through the API the recording is WAV, MP3, M4A, OGG, FLAC, AAC, WebM and Opus, up to 10 MiB.Cloning through the API
- Which engines can clone a voice, and in which languages?
- Five of the seven engines clone a voice from a recording: Qwen3-TTS (10 languages), LuxTTS (English only), Chatterbox Multilingual (23 languages), Chatterbox Turbo (English only) and TADA (10 languages). Kokoro and Qwen3-TTS CustomVoice speak built-in preset voices instead and do not take a recording. When you save a voice built from a recording, the studio offers only the engines that can clone it.The enginesEvery language
- Does voice cloning support Hindi or Marathi?
- Hindi yes, Marathi no. Hindi is one of the 23 languages VoiceLabs speaks, and Chatterbox Multilingual is the engine that both speaks Hindi and clones a voice from a recording. Kokoro also speaks Hindi, but only in its built-in preset voices. Marathi is not one of the 23 languages VoiceLabs accepts, so no engine here is set up for it. Hindi is written in Devanagari, which costs about four times as many model tokens per character as Latin script, so a long Hindi text is split into smaller pieces for you.Every language
- Does voice cloning run in the cloud or locally in my browser?
- In the cloud. Your recording is sent to VoiceLabs, the models run on our GPU servers, and the audio comes back to your browser tab or your code. Nothing is computed on your device, so there is no model to download and no graphics card is needed. The trade-off is that your recording and your text leave your machine; if they must not, a model you run yourself is the right choice.Cloud text to speechPrivacy policy
- Can I use my cloned voice for narration in my browser, through the API or in an AI assistant?
- Yes. The voice lives on your account, and every surface lists it. In the browser, the read-aloud extension for Edge and Chrome reads a page or a selection in it; a browser cannot add a voice of its own to its built-in read-aloud, so the extension asks VoiceLabs to make the speech and plays the result as audio. From code, the API and the @voicelabs/sdk package speak in it, and a key with the voice:clone scope can create one. In Claude, ChatGPT and Cursor, the MCP server speaks in it, and creates one once you approve “Create a cloned voice from a recording you provide” on its consent screen. The API and the extension need VoiceLabs Pro.The read-aloud extensionThe text to speech APIThe MCP server
- Is it allowed to clone someone else's voice?
- Only with their explicit permission, or where the material is otherwise legally authorized. The Responsible Use policy requires consent: by cloning a voice you affirm that it is your own, or that you have the speaker's permission. The studio asks you to agree to that before first use and records your acceptance, and the API reference says to clone only a voice whose owner has agreed to it. VoiceLabs cannot check whose voice a recording is, so the responsibility is yours. Impersonating someone without permission, fraud and scams are prohibited, and misuse can be reported.Responsible useWhat you may do with the audioReport misuse
- How much does voice cloning cost?
- Voice cloning is part of VoiceLabs Pro: US$8 a month or US$59 a year, with a 7-day trial that asks for a card up front. Pro includes unlimited saved cloned voices, all seven engines, API access and 2 hours of generated audio a month (fair use), and there is no per-use fee.Pricing
VoiceLabs Pro is US$8 a month or US$59 a year, with a 7-day trial. Voice cloning and the API are included.
VoiceLabs is built on the open-source VoiceBox project (MIT), and that attribution is permanent. The hosted service, with its GPU inference, accounts and billing, is a commercial product from Devino Solutions.