v2.5.5

Roughly half of Google Meet bots were falling back to on-screen observation, which attributes every speaker to one identity. Bots now hold the audio path that exposes each participant separately, and recover it if it drops mid-call.

⚡Improvements

• Google Meet transcripts attribute speech to individual participants instead of collapsing every speaker onto one. The bot keeps the per-participant audio path through the first seconds while the participant list resolves, rather than abandoning it before names arrive, and re-acquires it if it drops later in the meeting.
• Microsoft Teams attributes turn changes more accurately in meetings that emit both a dominant-speaker signal and live captions, where the previous speaker could be credited with the next person's words.

🐛Bug Fixes

• A bot no longer ends mid-meeting — losing the recording — when it cannot write its speaker timeline to disk. It keeps recording and writes the timeline at the end.

↩ All releases