ElevenLabs' New AI Transcribes Meetings Better, Even With Crosstalk
Fewer transcription errors mean less time fixing notes and more accurate records.
ElevenLabs launched Scribe v3 and Scribe v3 Live, Speech to Text models built for messy real-world audio like overlapping speakers, callers pausing mid-sentence, and conversations that switch languages. Scribe v3 makes 17% fewer diarization errors than Scribe v2 on recordings with three or more speakers, and Scribe v3 Live brings diarization to realtime transcription for the first time. Scribe v3 also reduces word error rate by 27% compared with Scribe v2 across 102 languages on FLEURS, with improvements in every language tested.
- Scribe v3 reduces transcription errors by 27% across 102 languages compared to the previous version.
- It's better at identifying speakers in group conversations, with 17% fewer mistakes when three or more people talk.
- The live version is only available in ElevenLabs' voice agents for now, not as a standalone app.
Why It Matters
More accurate transcripts mean less time correcting notes and fewer misunderstandings in meetings and calls.