TTSbox
Voice cloning for TTS Local browser processing No signup WAV download

Voice Cloning for Text to Speech

Provide a short authorized voice sample, enter text, and generate new speech in that voice directly in your browser. Only use voices you own or have explicit permission to clone.

1
2
3

Local Voice Model

Cached locally after first load

Only upload your own voice, or a voice you have explicit permission to use.

1
2
3

Guide

How to use voice cloning for text to speech

Workflow

How voice cloning works for text to speech

3 steps to clone a voice and generate speech locally.

View details
  1. Load Model: The AI voice model (~150 MB) downloads into your browser cache. One-time download per language.
  2. Choose Voice Source: Select "Clone a Voice" and upload an authorized 5–10 second clip, or record your own voice.
  3. Generate: Enter text (up to 1500 characters), and WebGPU runs inference natively to produce a WAV file.

This dedicated workflow only uses an uploaded or recorded reference voice. Kokoro built-in voices remain on the separate Text to Speech page.

Privacy

Local processing, no server upload

Your voice sample stays in your browser.

View details
  • No server upload of your voice sample or generated audio.
  • All processing runs locally via WebGPU on your device.
  • No account signup or API keys required.
  • Generated WAV files are saved directly to your device.

The model is downloaded from the internet on first use, then cached in your browser storage.

Voice Source

Provide an authorized reference voice

Upload a clip or record your own voice.

View details
  • Upload: Choose an authorized 5–10 second audio clip.
  • Record: Record your own voice directly in the browser.

Only clone voices you own or have explicit permission to use. See the Safety Policy.

Languages

Multilingual voice generation

Generate speech in 6 languages with dedicated AI models.

View details
  • English, French, German, Spanish, Portuguese, Italian
  • Each language requires its own model download (~150 MB).
  • Switch languages by loading the corresponding model.
  • Licensed sample voices available for each language.
Requirements

Technical requirements

What you need to run TTSBox voice cloning.

View details
  • Browser: Desktop Chrome or Edge (latest versions recommended).
  • API: WebGPU support enabled.
  • Hardware: A dedicated GPU or modern integrated GPU. Mobile browsers are not recommended.
  • Storage: ~150 MB downloaded to browser cache per language.
Use Cases

Who uses voice cloning for text to speech?

Who it is for

  • •Creators testing narration drafts before studio recording
  • •Developers exploring browser AI audio and WebGPU
  • •Researchers evaluating local inference efficiency
  • •Teams prototyping localized voice copy securely

Who it is not for

  • •Impersonation or deceptive media creation
  • •Fraud, scams, or phishing attempts
  • •Commercial use without proper voice licensing

FAQ

Frequently asked questions

Is voice cloning a separate tool from text to speech?
Yes. TTSBox keeps voice cloning separate from Kokoro text to speech. This page only accepts an authorized uploaded or recorded reference voice and does not display Kokoro built-in voices.
How does browser voice cloning work?
TTSBox loads a voice model (~150 MB) into your browser. You provide a short authorized voice sample, enter text, and the model generates speech in that voice — all locally on your device using WebGPU.
Do I need to create an account?
No. TTSBox is a free browser tool. No account, no sign-up, no server-side processing. Just open the page and start.
What languages are supported?
English, French, German, Spanish, Portuguese, and Italian are currently supported. Each language has its own model pack (~150 MB). The model must be loaded for the target language before generating.
How long should my voice sample be?
A clear voice sample of 5 to 10 seconds recorded in a quiet environment works best. The maximum allowed length is 15 seconds.
Can I use this on mobile?
Mobile browsers are not recommended. Browser-based voice cloning requires WebGPU support and significant GPU memory, which most mobile devices currently lack or restrict. Use a desktop browser with Chrome or Edge for the best experience.
How is this different from cloud voice cloning?
Local browser voice cloning runs inference on your device using WebGPU, meaning your voice sample and generated audio are not uploaded to a server. Cloud voice cloning processes audio on remote servers, which may be faster on weak hardware but requires uploading your voice data to a third party.