ElevenLabs launches Scribe v2 Realtime, its latest ultra-low-latency speech-to-text model

ElevenLabs has released Scribe v2 Realtime, its latest low-latency speech-to-text model for real-time transcription. It processes speech in about 150 milliseconds, making it ideal for live applications that need instant conversion. The company claims it sets a new standard for accuracy among real-time ASR systems, performing especially well on difficult audio with background noise or complex content.

The model supports more than 90 languages, including English, French, German, Italian, Spanish, Portuguese, Hindi, and Japanese. It is optimized for use in voice agents, meeting transcription, and conversational AI, extending ElevenLabs’ tools for real-time customer interaction and automation.

Scribe v2 Realtime meets enterprise security and privacy standards such as SOC 2, ISO27001, PCI DSS Level 1, HIPAA, and GDPR. It also offers EU and India data residency options and a zero-retention mode. The model can be used through ElevenLabs’ API or within ElevenLabs Agents, with enterprise plans supporting 30 or more concurrent sessions.