November 2026
Whisper Memos now uses ElevenLabs Scribe v3, with 27% fewer word errors and clearer speaker labels in meetings.
Whisper Memos now runs on ElevenLabs Scribe v3, the most accurate speech to text model ElevenLabs has made. It makes 27% fewer word errors than Scribe v2 across 102 languages, and it improved in every language ElevenLabs tested.
Some languages got a lot better: German has 42% fewer errors, Spanish 35%, and English 23%. Harder languages like Swahili, Tamil, and Thai improved by around 30%. In a long recording, that means far fewer words you have to go back and fix.
Scribe is the model behind Meeting mode, so this is where you'll notice it most. Scribe v3 has the lowest speaker error of any model ElevenLabs tested. It mixes up speakers 12% less often than v2, and 17% less often when three or more people are talking.
Quick back-and-forth, people cutting in, a third voice joining the call: your transcript now keeps track of who said what.
Scribe v3 is already live. If you use ElevenLabs Scribe in Settings → AI Model, or record meetings in Automatic mode, your next memo uses it. Your custom words keep working too, so names and jargon stay spelled the way you expect.
Compare all AI models to see which one fits your recordings best.
