A Guth Labs publication ↗
Back to News

ElevenLabs launches Scribe v3 and Live with new speech-recognition controls

ElevenLabs launched Scribe v3 for recordings and Scribe v3 Live for streaming audio on November 10. Both are available through its API, with Live also offered in ElevenAgents. The streaming model adds speaker labeling and conversation-aware turn detection designed to distinguish a pause from the end of a speaker’s turn. Launch announcement

The company reports 27% fewer word errors for recorded-audio Scribe v3 than Scribe v2 across its 102-language FLEURS evaluation. Those results are company measurements. Its separate speaker-labeling evaluation uses an internal collection of meetings, calls and voice notes; results from differently scored datasets should not be treated as interchangeable. Evaluation method

The documentation lists output controls including digits-versus-words formatting, mixed-language handling and terminology hints. The API identifiers are scribe_v3 and scribe_v3_live. These are speech-recognition models, distinct from ElevenLabs’ speech-generation products. Model documentation

For developers, the change is most relevant where a transcript triggers another system’s actions. Speaker attribution, numbers and interruption timing can matter as much as overall transcription accuracy. Testing on an application’s actual accents, background noise and terminology remains more useful than treating a headline benchmark improvement as a deployment guarantee. No transcription tests were performed for this report.

Sources

Sources and citations

Revision history

Published .