xAI launches custom voices for Grok

xAI now lets you clone your voice with approximately one minute of audio and use it in Grok to create conversations, narrations, and customer service agents. The feature, called Custom Voices, is available through the company’s text-to-speech and voice agent APIs.

The process takes less than two minutes. You record a natural voice sample in the xAI console, read a verification sentence, and the system generates a production-ready model. You can then use that voice anywhere xAI offers its built-in voices.

What you can do with your cloned voice

The feature is designed for developers and businesses that want to avoid the generic voices used by text-to-speech systems. For example, a retailer can create a customer service agent that responds with a recognizable voice consistent with its brand.

You can also:

• Narrate videos, podcasts, and social media posts without rerecording every piece of text.
• Create complete audiobooks, chapter by chapter, using the author’s or narrator’s voice.
• Generate dialogue for characters without booking studio time for every line.
• Produce messages from a company or CEO in multiple languages.
• Create a custom voice for people who have lost the ability to speak while preserving their vocal identity.

Custom voices retain the capabilities of xAI’s text-to-speech system. That includes tags for controlling delivery, multilingual output, and real-time streaming through REST or WebSocket connections, which are ways for an application to send and receive data.

The same voice can also be used with xAI’s voice agent API. In practice, a developer can assign a voice_id to an agent so it can hold real-time conversations with a specific voice.

Cloning requires proving that it is your voice

xAI says you cannot create a custom voice from an old recording or clone someone else’s voice. Each request goes through a two-step verification process.

First, the person must read a sentence provided by the system. The speech recognition engine transcribes what is said and compares it in real time with the expected text. This confirms that the person is present and has consented to creating the voice.

Next, xAI compares the mathematical characteristics of the voice in that sentence with those of the full recording. These representations, known as speaker embeddings, confirm that both samples belong to the same person.

“You cannot clone a voice from a preexisting recording, and you cannot clone someone else’s voice.”

What changes for you

If you are a creator, you can turn scripts into audio without returning to the studio every time you change a line. If you develop software, you can offer voice agents with a more recognizable identity. And if a company operates in multiple countries, it can generate versions in English, Spanish, French, German, Chinese, Japanese, and other languages while preserving the same reference voice.

xAI says it will not charge an additional fee for using custom voices with its text-to-speech or voice agent APIs. The general cost of those services, where applicable based on usage, remains a separate matter.

The point to watch is how this protection works when cloned voices are used at scale. Presence and ownership verification reduces the most obvious risk, but it does not by itself eliminate impersonation, fraud, or misuse of generated audio.