Voice fingerprints
Name once.
Recognized forever.
Cloud notetakers re-label the same coworker as Speaker 1, Speaker 2, Speaker 3 across every call. Mac Note Taker remembers their voice - locally, with CAM++ embeddings - and auto-fills the right name from then on.
Anya
PM · 14 meetings tracked
“Tuesday works. Marketing site can ship Monday night.”
Marko
Eng · 9 meetings tracked
“Founder coupon copy is in. Drop it on the hero too?”
Lina
Design · 22 meetings tracked
“I'll redo the OG image with the new palette.”
Theo
Sales · 6 meetings tracked
“Demo went well; they want pricing by Thursday.”
Priya
Founder · 31 meetings tracked
“Let's lock the launch date - Tuesday or Thursday.”
Owen
youyou · 31 meetings tracked
“Tuesday. I'll cut the release branch tonight.”
Across meetings
Same voice. Same name.
You renamed Priya once, on May 2. Every meeting after that auto-labels her. The voice fingerprint database is on your Mac - leaves nothing.
How the matching works
Diarize
pyannote-segmentation-3.0 splits each meeting's audio into per-speaker turns. CoreML on the Apple Neural Engine, real-time.
Embed
CAM++ converts each turn into a 192-dim voice fingerprint. The fingerprint is unique per voice but doesn't include any audio.
Match
New fingerprints get cosine-similarity-matched against your stored library. Above threshold (0.65) → known speaker. Below → new speaker.
You own the voice library.
Forget a voice
Settings → Speakers → Forget. Removes the fingerprint. New recordings won't match them.
Rename anywhere
Renaming Priya in any meeting renames her in every meeting that already labeled her.
No name leaves the Mac
Speaker names live in your local SwiftData store. They never reach our servers.
Color-coded by default
Each speaker gets a stable color across meetings - quick visual scan in long transcripts.
Stop re-labeling the same coworker.
$149 lifetime · 3 Macs · code FOUDNER for $79 (first 100).