Skip to content

ElevenLabs Releases Scribe v3 and Scribe v3 Live

ElevenLabs

The speech-to-text models support 102 languages and add real-time speaker diarization and context-aware turn detection for voice agents.

Key points

  • Scribe v3 reduces word error rate by 27% and makes 17% fewer diarization errors on recordings with three or more speakers.
  • Scribe v3 Live cuts semantic word error rate by 43% compared to Scribe v2 Realtime.
  • Turn detection evaluates conversational context rather than audio silence to prevent interrupting callers who pause mid-sentence.
  • Prompting supports plain English formatting instructions, keyterm lists up to 1,000 words, and custom vocabulary libraries holding up to 10,000 terms.
  • Both models are available in ElevenAPI and ElevenAgents, with enterprise data residency options in the US, EU, India, and Singapore.

Why it matters

Developers building voice agents and transcription tools get improved multi-speaker labeling and streaming accuracy without tuning silence thresholds.

Sources