Intel / article

Scribe V2 Just Got An Upgrade

Key takeaways · AI-distilled

Scribe v2 now redacts names, card numbers, Social Security numbers and other PII during transcription, as [REDACTED], an entity type such as [CREDIT_CARD], or a numbered label like [CREDIT_CARD_1]. ElevenLabs warns it may miss some data and outputs need review.
For Indic-English code-switching, Scribe v2 keeps English words in Latin script whether the language code is English, Hindi or omitted, and this covers pairs such as Telugu-English and Kannada-English, not only Hindi.
A new No Verbatim mode removes fillers such as "um" and "uh", repeated phrases and stuttering, aimed at meeting notes and subtitles rather than raw capture of every sound.
Keyterm prompting rises from 100 to 1,000 terms per transcript and stays -aware, using surrounding audio to decide whether a term applies, but requests with more than 100 keyterms carry a 20-second minimum billable unit.

Terms in this piece · Glossary

context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.