ElevenLabs Launches Scribe v3 Speech Recognition Models
ElevenLabs released the Scribe v3 and Scribe v3 Live speech recognition models, covering 102 languages. The company says they cut interrupted speech far more than before.
A voice AI company called ElevenLabs launched two new automatic speech recognition (ASR) systems, Scribe v3 and Scribe v3 Live. They’re able to transcribe audio in 102 different languages. According to ElevenLabs, Scribe v3 reduces transcription errors by 27% over their prior ASR system, while Scribe v3 Live reduces word error rate by 31% and semantic error rate by 43%.
Scribe v3 Live is designed to address the frustration of being interrupted by a voice assistant when you take a pause during your sentence — the AI mistakenly thinks you’ve finished speaking. It also claims to understand the context of a conversation to determine when a speaker has stopped talking, rather than just relying on periods of silence. This could be useful for businesses that use an automated voice assistant to field customer service calls.
These results were achieved by comparing ElevenLabs’ new ASR systems to their old ones, not other major players such as OpenAI, Google, or Microsoft. Only time will tell if customers experience less interruption and if these improvements hold true in practice.
Source: elevenlabs.io
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles