Scribe adds real-time speaker labels for messy group calls

What changed

• The transcription model was retrained on overlapping, cross-talking speech so it can attribute lines correctly even when two people talk at once.
• Speaker labels now update live in the transcript pane instead of only after the call ends.
• A 'who said what' summary is generated automatically at the end of the call, grouped by speaker.

Who it helps

The practical problem it solves

Group calls with cross-talk produce transcripts that mix up who said what, which makes the notes almost useless for anything beyond a rough summary. Reliable live speaker attribution turns a chaotic call into a transcript people can actually cite and act on.

Ways to try it

• Run a 6-person usability session and get an accurate per-participant quote log without manual cleanup.
• Turn a cross-team standup into assignable action items grouped by who committed to them.
• Skip a call and read the 'who said what' summary in under two minutes.

More from this briefing

• Inbox Copilot now drafts replies from your team's past threads

  Coding agents

• Kestrel Search adds a 'show your work' citation mode

  Search & retrieval