Minutiva Private v0.6.0-beta4
Three fixes in this build target the most likely reasons earlier betas felt “mysterious” mid-session — plus an instant first-recording start.
What’s fixed
Recording-start errors are now visible. If starting a recording fails — most commonly because your input device is in Bluetooth voice mode (AirPods just off a FaceTime/Slack/Meet call) — the “Before recording” window now stays open and shows what went wrong, with an actionable hint:
“Your input device is in Bluetooth voice mode (probably AirPods on a call). Open System Settings → Sound → Output and switch to MacBook Pro Speakers, then try again.”
Previously the window would just vanish, leaving you wondering whether the recording started or not. (Spoiler: it hadn’t.) This was the single most-common silent failure mode in beta3.
No more crash on the third recording-in-a-session.
Switching the elapsed-time counter in the floating recording window from a Combine timer subscription to SwiftUI’s TimelineView eliminates a Swift 6 main-actor isolation race that could SIGSEGV when the view tore down and rebuilt across consecutive recordings. Record, stop, record, stop, record — safely.
What’s faster
First recording starts ~10x sooner. The Whisper transcription model (~3 GB for Large V3) now loads in the background as soon as Minutiva launches, instead of on the first click. By the time you hit New recording, it’s ready — no 10–15-second wait while the model warms up.
New: keep the model loaded between recordings. For people who do many recordings in a session, there’s a new opt-in toggle at Settings → Transcription → Performance → “Keep transcription model loaded between recordings.” With it ON, subsequent recordings start instantly (no model reload). Off by default to be friendly to 16 GB Macs; recommended on for 24 GB+.
Also in this build
- People diagnostics + bulk re-match —
Persondetail page shows the top-5 closest unmatched speaker clusters across your meetings (cosine similarity bars, green when above suggestion threshold). Settings → Advanced → “Re-run People matching across all meetings” refreshes suggestions library-wide without re-clustering. - Live speaker labels during recording (Premium, live mode). While recording, the transcript pane now shows preliminary speaker labels (Speaker A/B/C) refreshed every 30 seconds via a periodic VBxHMM snapshot. Labels may shift across snapshots until you stop — the final clustering happens at stop as before.
Nothing changed in
- VBxHMM clustering algorithm + merge-back threshold
- People model and file formats
- License code format — existing beta keys keep working
- Audio capture pipeline (Bug C/D/E stayed fixed)
Auto-update
Existing beta channel testers get this within 24 hours. Or trigger now: Minutiva Private → Check for Updates…