Skip to main content
Grok Voice Transcribe 2 is xAI’s current speech-to-text model. It transcribes 38+ languages and is notable for two features that matter in production media work: multichannel input, where each recorded channel is transcribed separately, and keyterm prompting for names and terminology.
Available in Ollang Workflows as grok-voice-transcribe — Closed Captions, Subtitle Translation, and AI Dubbing · New. See the Speech-to-Text catalog.

Model and coverage

xAI retired Grok Voice Transcribe 1.0 in October 2026; requests to it are now served by 2.0.

Key capabilities

  • Multichannel transcription. When each speaker was recorded on their own channel, each channel is transcribed on its own, which avoids the attribution errors that come from separating mixed speech.
  • Speaker diarization with a speaker label on every word.
  • Word-level timestamps, which give subtitle segmentation precise boundaries.
  • Keyterm prompting to enforce brand, product, and people names.

Where it fits in a workflow

Grok Voice Transcribe 2 suits podcasts, interviews, and panel recordings with a channel per speaker, and content dense with names that a general model would misspell. For languages outside its list, use AssemblyAI or ElevenLabs Scribe v2; for genuinely difficult mixed audio, use Agentic STT.

Trade-offs

  • 38+ languages is narrower than AssemblyAI (99), ElevenLabs Scribe v2 (90+), or Speechmatics (70+). Confirm your source language first.
  • Formatting coverage is partial: automatic text formatting applies to 11 of its languages.
  • Newer model line, with less production history than the established providers.

Reference