ElevenLabs Launches Scribe v3 Speech-to-Text Models

At a glance

ElevenLabs has released two new speech-to-text models, Scribe v3 and the real-time Scribe v3 Live. They are designed to transcribe accurately even when several people talk at once or speech is interrupted.

Key points

1. 1

   Scribe v3 makes 27% fewer word errors than Scribe v2 on recorded audio and supports 102 languages.

2. 2

   Scribe v3 Live makes 43% fewer errors than Scribe v2 Realtime on live audio.

3. 3

   Scribe v3 Live includes speaker identification, which tells who said what, for the first time.

4. 4

   Both models are available through ElevenAPI, and switching from Scribe v2 only requires changing the model ID.

Why it matters

Speech recognition underpins meeting notes, contact centers, and voice agents, so better accuracy improves the quality of all these services.