What's New
Multi-speaker diarization (#120, PR #130)
OpenOats can now distinguish multiple remote speakers on system audio using FluidAudio's LS-EEND diarizer. When enabled in Settings → Speaker Diarization:
- Streaming: Live transcription labels each segment with the dominant speaker (Speaker 1, Speaker 2, etc.)
- Batch refinement: Post-meeting re-transcription runs full offline diarization for accurate speaker attribution
- Smart fallback: Single-speaker calls show "Them" as before — no visual change unless multiple speakers are detected
- 10-color palette: Each speaker gets a distinct color in the transcript view
- Dynamic participants: Markdown meeting notes list actual speakers detected
Three model variants are available:
- DIHARD III (default) — general-purpose, up to 10 speakers
- AMI — optimized for in-person meetings, up to 4 speakers
- CALLHOME — optimized for phone calls, up to 7 speakers
Dependency changes
- FluidAudio bumped from 0.7.9 to 0.12.5
- WhisperKit sourced from fork with relaxed swift-transformers constraint (builds verified with swift-transformers 1.2.0)
Homebrew
brew tap yazinsai/openoats https://github.com/yazinsai/OpenOats
brew install --cask yazinsai/openoats/openoats