ElevenLabs Enhances Speech-to-Text with Scribe v3 and Realtime Diarization

ElevenLabs launched Scribe v3 and Scribe v3 Live on November 10, 2026, introducing its most accurate speech-to-text models to date with significant improvements for both recorded and real-time audio transcription across 102 languages. For broader context, explore our Top 100 AI Tools.

Improved Accuracy for Recorded Audio with Scribe v3

Scribe v3, designed for recorded audio, demonstrates a 27% reduction in word errors compared to its predecessor, Scribe v2, across 102 languages. This model also features the most accurate speaker labels among all models tested by ElevenLabs, improving the clarity and usability of transcripts. Its speaker diarization error rate is 12% lower than Scribe v2 overall, and 17% lower for recordings with three or more speakers.

Real-Time Transcription Advances with Scribe v3 Live

For real-time applications, Scribe v3 Live offers a 43% reduction in errors compared to Scribe v2 Realtime. A notable new feature is the introduction of speaker diarization to real-time transcription, with a 3.1% real-time diarization error rate observed in English. Scribe v3 Live also incorporates conversation context for turn detection, which helps prevent premature cut-offs when callers pause during a conversation.

Developer and Enterprise Features

ElevenLabs has integrated several features into Scribe v3 and Scribe v3 Live to support developers and enterprise clients. Developers can utilize keyterm prompting with up to 1,000 terms, a keyterm library, and plain-English transcript formatting rules. A multilingual mode is also available. For enterprise users, the models are HIPAA eligible and offer a zero-retention mode. Data residency options are provided in the EU, India, and Singapore, alongside certifications such as SOC 2, ISO 27001, and PCI DSS Level 1.

Availability and Impact

Both Scribe v3 and Scribe v3 Live are accessible via ElevenAPI and the ElevenLabs app. Scribe v3 Live is additionally available within ElevenAgents. ElevenLabs reports that its Scribe technology has been used for hundreds of millions of hours of transcription across various applications, indicating its widespread adoption in the speech-to-text domain.

Conclusion

The launch of Scribe v3 and Scribe v3 Live by ElevenLabs marks a significant update in speech-to-text technology, offering enhanced accuracy and new features for both recorded and real-time audio transcription. These improvements aim to provide more reliable and context-aware transcription services for a broad range of users, from individual developers to large enterprises.

Sources

• GitHub - KORThomasJeong/live-scribe: Real-time meeting transcription in an Obsidian sidebar, powered by ElevenLabs Scribe · GitHub
• stt-benchmarks/README.md at main
• Optional ElevenLabs voice provider: Scribe v2 Realtime dictation and live voice mode via Agents · Issue #1844 · nightly-labs/openbot · GitHub
• skills/speech-to-text/SKILL.md at main · elevenlabs/skills · GitHub
• Laying the Open-Source Foundation of Unified-Dialectal...