xAI introduces custom voice cloning with two-stage safety verification
Developers can now create custom voice models in under two minutes by recording natural speech directly in the xAI console.
2 min read·Friday, 16 October
xAI has launched a new custom voice cloning feature that allows developers to create production-ready voice models in under two minutes.
Key facts
Creation time
Under two minutes
Required recording
About a minute
Creating a Custom Voice
To build a custom voice, users record about a minute of natural speech inside the xAI console. The processing pipeline verifies ownership, processes the recording, and generates a usable voice model rapidly. Once created, these custom voices inherit all standard text-to-speech capabilities, including speech tags, multilingual output, and support for both REST and WebSocket streaming.
Custom voices function seamlessly wherever built-in voices are used. Developers simply pass a specific voice identifier to any text-to-speech endpoint or integrate it with the Voice Agent API for real-time conversational agents. xAI notes that there is no extra charge for using custom voices with these APIs.
Voice Safety and Verification
To prevent misuse, every custom voice must pass a two-stage verification process before creation.
First, the speaker must read a specific verification phrase. An automatic speech recognition engine transcribes and matches this phrase in real time to confirm intent and presence. Second, the system computes speaker embeddings — mathematical representations of vocal characteristics — from both the verification clip and the full recording to ensure they belong to the exact same person.
Because of these safeguards, users cannot clone a voice using pre-existing recordings, nor can they clone someone else's voice without their live participation.
Comments