The Grok team introduced Voice Transcribe 2.0, emphasizing noisy, multilingual speech and conversational recordings. The announcement includes batch and streaming use, word-level timestamps, speaker labeling, and vocabulary hints, alongside the company’s accuracy evaluations.
Context
Real-world audio contains names, interruptions, and short phrases with little context. Those cases can matter more than clean narration. Teams should test representative recordings and inspect whether uncertainty is preserved for review, particularly when transcripts feed search, summaries, or downstream automated decisions.
Sources & authors
- Introducing Grok Voice Transcribe 2.0SpaceXAI / xAI · September 18, 2026



