ElevenLabs ships Scribe v3: 43% fewer real-time errors, 102 languages

minchoi · x · 2026-11-11

ElevenLabs released Scribe v3 and the real-time Scribe v3 Live, its most accurate speech-to-text models yet, built for real-world audio:

43% fewer meaning-changing errors in real-time transcription vs the previous version; it can tell a pause from the end of a sentence, so voice assistants stop interrupting mid-thought.
Tracks who said what even with multiple speakers in noisy meetings.
Supports 102 languages with better in-conversation language switching.
Ships 10,000 specialized terms covering medical, legal and product vocabulary.
Edit transcripts with plain-English instructions for dates, numbers and formatting.

Already in production, e.g. at a Dutch energy company.

More from Multimodal