Capture Meet live captions → real-time transcript (no audio/STT)

[YJack0000]

Goal

Get a real-time transcript for free by scraping Google Meet's live captions instead of capturing audio — feeding Parley's local analysis. This keeps the feature zero-cloud and avoids any STT cost for joined/observed meetings.

Tasks

• Programmatically enable live captions after joining.
• MutationObserver on the caption DOM → { speaker, text, start, end } segments; dedupe partial → final.
• Stream segments to Parley over localhost (Tauri command / websocket), tagged with their meeting/session id.
• Resilience: handle captions unavailable, DOM/selector changes, and language settings; clear fallback + logging.

Notes

• Meet captions carry speaker names → accurate attribution for free (better than guessing via diarization).
• Tradeoffs: ~90% accuracy, misses heavy cross-talk, language-dependent, no audio recording. Acceptable for live coaching / catch-up.

Part of the "Multi-meeting co-pilot & catch-up mode" epic.