Speakers
How names get onto transcript lines: the call window, voices, people in the room, names said aloud, and your corrections.
The Speakers page decides how PrivateGoat works out who said each line. All of it runs on this Mac, and a transcript without names is still complete.
Recognise people by voice
The switch at the top of the page, under By voice. It turns the voice features on or off together: naming by voice, people in the room and — where the vault keeps them — remembered voices and learning your own. The rows below still change one at a time, and turning it off deletes nothing. Off, names come only from the call window. The voices stay on this Mac.
Names from the call window
On by default. In Teams, Zoom's gallery view and Google Meet, the tile marked as the person speaking is read on the Mac and its name goes onto their lines. It needs Save screenshots of the call window on Calls, switched on before the meeting. Other call apps give no names this way.
Names by voice
Name call speakers by voice, on by default under Recognise people by voice. A small voice model groups the voices the call window could not name and fills only lines nobody named, so it never overrules the window. It needs a one-time download of about 22 MB.
Remembered voices
Remember voices across meetings is off by default. It keeps the voice of each person the call window named — only inside the encrypted vault, so it cannot be turned on until the vault keeps voice profiles. Name people by their remembered voice then names voices the window could not, such as someone on the phone. The call window's names always win. Forget all voices shows how many are remembered and removes them all, your own included; the People page forgets one at a time.
People in the room
Tell apart people in the room is off by default; Quick setup turns it on once the vault is on. It gives colleagues at your table their own lines instead of yours. A line stays yours unless the voices clearly differ. It learns your voice from meetings where you were alone at the microphone, kept only in the vault; Forget My Voice… removes it.
Echo reduction
On together with people in the room. Without headphones, the microphone also records what your speakers play. Echo reduction plays each voice only once when you listen back.
Fold lines heard twice
On by default. When the microphone and the call both caught the same sentence, the transcript shows it once, with the copy folded underneath. Summaries, exports, search and coding agents read it once. When the words differ too much to be sure, nothing is folded.
Names people say
Use names people say in the meeting is on by default. A greeting ("Good morning, Dana"), a hand-over or a thank-you can name the voice that answers. A spoken name only counts with a second cue or the call window's agreement, never against the call window, and the line says why it was named. It applies when a meeting is named again.
Name corrections
A name you correct with Use this name in every meeting becomes a rule, listed here as "shown name → your name". Remove one to show the call window's own name again in every meeting.
For the curious
- Every named line explains in its tooltip why it was named. A voice named from a clear lead in the call window's votes is marked as not sure.
- The voice model and the voice groups run on the Mac. Remembered voices live in
…/Sonavia/Voices/, speaker pictures in…/Sonavia/People/. - Keep a picture of each speaker (off by default) crops each person's tile from the call screenshots. Delete Speaker Pictures… and Delete Voice Profiles… remove them all at once.
- Name Speakers Again on a meeting reads its pictures and voices again from scratch.
What leaves your Mac
Nothing, except the one-time voice model download from Hugging Face, the same host as the speech models. Names, pictures and voices stay on this Mac.
Related
- People and speaker names — the People page, renaming and merging.
- Call screenshots — the pictures names are read from.
- Playback and corrections — fixing words and names.