AI BriefingElevenLabs Unveils Real-Time Transcription Model 'Scribe V2 Realtime'
Key point
ElevenLabs has unveiled **Scribe V2 Realtime**, a real-time transcription model with ultra-low latency performance under 150ms.
Details
Scribe V2 Realtime is an ultra-low-latency Speech to Text (STT) model designed for real-time use cases. It is optimized for voice agents, meeting assistants, and real-time caption generation, delivering latency under 150ms across 90 languages.
It is especially designed to specialize in agentic use cases. In tests on 500 challenging samples containing background noise and complex information, it overwhelmingly outperformed existing models, achieving 93.5% accuracy across 30 major European and Asian languages.
Key features include:
- Negative latency: predicts the next word and punctuation
- Automatic language detection: supports language switching mid-conversation
- Text conditioning: supports continuous transcription based on previous batches
- VAD (Voice Activity Detection) and Manual commit features
- Enterprise-grade security: compliance with SOC 2, ISO 27001, HIPAA, and more
The model is now available immediately through the ElevenLabs API and ElevenLabs Agents.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Comfy Launches Comfy Router: A Unified API for Frontier Media Models