Grok Text to Speech

Grok Voice Clone: How xAI Custom Voices Work and How to Make One

xAI’s Custom Voices API clones a voice from about two minutes of audio with a two-step verification, then lets you use it in Grok text to speech and voice agents. Requirements, steps, cost and limits.

Last updated:

Grok voice clone is the informal name for xAI's Custom Voices feature: you give the Grok Voice API about two minutes of your own speech, prove the voice is yours, and get a voice_id you can use exactly like Ara, Eve or any of the 28 built-in voices, in text to speech and in voice agents.

This page explains what it is, what it needs, how to do it and what it costs, from xAI's public documentation and Voice API page as of 16 October 2026. This site does not clone voices itself; it only lets you try the built-in ones.

What you get

A custom voice behaves like a built-in one:

  • Pass its id as voice_id to POST /v1/tts and you get your voice reading any text.
  • Use the same id as the voice of a Grok Voice 3.0 speech-to-speech agent.
  • Everything the built-in voices can do carries over: speech tags ([pause], [laugh], <whisper>…), all 20 validated languages, streaming, timestamps, every output format.

What xAI requires

  • About two minutes of natural speech. Clean audio, one speaker, normal conversational delivery. xAI's page says "clone from two minutes of natural speech".
  • Two-stage verification. You read a passphrase, and xAI matches a speaker embedding between the passphrase recording and the reference clip. Both must agree. This is what stops the feature being used on other people's voices.
  • An xAI account at console.x.ai.

How to clone a voice

In the console (no code). Go to Voice → Voice library, choose to create a custom voice, record or upload the reference audio, complete the passphrase check, and save. Open the voice card's menu and pick Copy Voice ID. xAI's docs note you can create a custom voice for free this way.

With the API. The Custom Voices API (docs) exposes the same flow for automation, and GET /v1/custom-voices lists the voices on your account so your app can offer them to users.

Use it.

curl -X POST https://api.x.ai/v1/tts \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Hello! This is my custom voice.", "voice_id": "YOUR_VOICE_ID", "language": "en"}' \
  --output hello.mp3

Cost

StepPrice
Creating the voice in the consoleFree, per xAI's docs
Text to speech with the voice$15 per 1M characters, same as built-in voices
Using the voice in a speech-to-speech agent$0.08 per minute, same as built-in voices

Full price list and calculator on the pricing page.

Tips for a better clone

  • Record in a quiet room with the microphone a hand's width away; avoid music beds or reverb.
  • Speak the way you want the clone to sound. If the reference is flat, the clone will be flat.
  • Cover a range: a few questions, a few statements, some longer sentences, so the model hears your intonation patterns.
  • After cloning, test with speech tags: <soft> and <emphasis> tend to show whether the character of the voice came through.

Limits and ethics

Verification means you can only clone yourself (or someone who records the passphrase with you present and consenting). That rules out impersonation of public figures, voices from videos, and "voice of a deceased relative" requests, which other services have been criticized for. If you need a brand voice that is not a real person, build it with the built-in voices and speech tags instead; try combinations on the homepage.

Questions