Speakers

Speaker chips on a transcript

When diarization is on, each detected speaker appears as a chip at the top of the transcript, showing their name and, when available, a confidence percentage. Opening a chip's menu gives you:

• Show only this speaker — filters the transcript down to just their segments.
• View Profile — jumps to that person's page in the Speaker Manager (only for chips backed by a saved voice profile).
• Select All Segments — bulk-selects everything attributed to them, so you can reassign the whole selection to someone else in one action instead of correcting line by line.
• Change speaker… — reassigns this speaker's segments to someone else.
• Merge With… — combines this speaker with another.
• Edit Voice Label — renames the speaker. If the chip has a saved voice profile, this renames it everywhere it's used, not just in this transcript.
• "«Name» is not in this call" — removes a wrongly detected speaker; every segment they were credited with is automatically re-attributed among the remaining people, nothing is deleted. This works both while a recording is still in progress and after it's finished.

A + button next to the chips lets you add a person the app didn't detect as their own speaker — every currently-unresolved segment moves onto them at once.

Placeholder

The "Who said this?" picker

This is the shared dialog for changing or adding a speaker anywhere in the app. It shows the current speaker first, then suggested matches, then people already seen in this transcript, then the rest of your voice library with a similarity percentage — typing a name searches everyone, not just the filtered shortlist. A pinned footer offers New person…, Not sure / Unknown, and Doesn't matter (for background noise or music that doesn't need attribution).

The Speaker Manager

The Speaker Manager is a full-screen library view for your saved voices, independent of any single recording: rename, recolor, merge duplicate entries, and review a voice's quality.

• Merging — select two or more speakers to combine them into one profile; with exactly two selected, a similarity percentage helps you judge whether the merge makes sense.
• Health and Deep Clean — a plain-language readout ("sounds like one voice" or "looks like N people") flags a profile that may have picked up more than one person's voice, with a guided Review & Split or a more thorough Deep Clean flow to sort it out.
• Deep Clean's audio playback expires with the same 72-hour retention window as any other recording — after that, the play button for a cluster is disabled, but the underlying voice data and transcript text keep working, since neither of those expires.

The advanced tools on this screen — the quality health readout, the contamination-sensitivity slider, Review & Split, and Deep Clean — apply to on-device speaker profiles; there is no cloud equivalent for them. If you're signed in with sync on, your speaker profiles themselves sync with your account, but this decontamination toolkit is local only.