Grok Voice Transcribe 2.0 focuses on noisy multilingual speech
The September speech-to-text release keeps the prior pricing and adds an updated accuracy claim.
THE STORY IN THREE POINTS
- Transcribe 2.0 launched on September 18.
- It supports batch and streaming transcription.
- The launch includes timestamps, diarization and domain-term biasing.
The announcement
The Grok team released Voice Transcribe 2.0 on September 18. The announcement describes batch and streaming transcription, speaker labels, word timestamps and support for language changes within a recording. It lists unchanged pricing of $0.10 per audio hour for batch and $0.20 for streaming. The company reports improved performance on noisy and multilingual audio; my ai daily has not reproduced those accuracy comparisons.
Editorial context
A transcript’s usefulness depends on which mistakes it makes. An incorrect filler word and an incorrect account number have very different consequences. A meaningful evaluation would include names, numbers and specialized vocabulary from the intended workflow, while also checking whether words are assigned to the correct speaker. Average word error alone can hide those distinctions.
Editorial context: what to watch
Teams could compare outputs on recordings whose correct transcripts are already known, preserving accents and background conditions. Human correction time would be a useful companion to automated scores. If transcripts feed another agent, the evaluation should also measure whether recognition errors change its final action, because a small transcription mistake can propagate beyond the text itself.
Sources: Official Grok Voice Transcribe 2.0 announcement, September 18 ↗
