Speaker Identification

Cadeo automatically identifies different speakers in your recordings, labeling transcript segments so you know who said what.

Automatic Identification

Speaker identification works differently by platform:

Platform	Method	Quality
Desktop (sherpa-onnx)	Native diarization	Best — identifies speakers automatically
Web (AssemblyAI cloud)	Cloud diarization	Excellent — included with cloud transcription
Web (local Whisper)	Basic separation	Limited — depends on audio quality

Matching Speakers to Attendees

For meetings linked to calendar events with attendees, a Match Speakers to Attendees card appears above the transcript.

1. Each speaker label (e.g., "Speaker 0", "Speaker 1") has a dropdown
2. Select the matching attendee name from the event's attendee list
3. Click Apply Matches to update the transcript with real names

The card can be dismissed and won't reappear for that recording.

Voice Fingerprinting

Cadeo learns each person's voice the first time you match them, so the next meeting you record with the same people picks the right names automatically — no card, no dropdowns.

How it works:

• The first time you confirm "this speaker is Alex", Cadeo saves a private fingerprint of Alex's voice alongside their name.
• On every future recording, each speaker's voice is compared to your saved fingerprints. Confident matches are applied automatically.
• If Cadeo's not sure, the Match Speakers to Attendees card still appears so you can confirm.
• Every confirmation makes the fingerprint a little stronger, so matches keep improving.

Voice fingerprints are stored privately in your workspace. They're never shared between workspaces and never sent to any third party for matching — comparison happens directly against your own data.

If a fingerprint goes wrong (e.g. a recording was mislabelled), the next correct confirmation overwrites it.

Fixing Who Said What

When people talk over each other, two voices can end up under one speaker. You can fix that by hand, a passage at a time.

Click a speaker's name on any paragraph of the transcript. At the top of the box that opens, choose what the change is for:

• Everywhere — renames this speaker in the whole transcript (as before). Naming someone this way also teaches Cadeo their voice.
• This passage — moves just this paragraph to someone else. Pick a name from the Move to list (typing filters it; ↑ ↓ and Enter pick from the keyboard), or type a new one.
• Part of it… — for a paragraph that holds two people. Put the cursor where the other person starts and press Enter: the paragraph splits there, and you pick who said the second half. Esc cancels.

You can also select some words in a paragraph and click Change speaker… to move exactly those words.

After each change a message offers Undo, which puts the paragraph back as it was. The summary and action items update a few seconds later.

Moving a passage never changes anyone's saved voice, so a paragraph that mixed two people can't teach Cadeo the wrong voice.

Transcript Display

Each transcript segment shows:

• Speaker name (or label if not yet matched)
• Timestamp
• Text content

Following along

While a recording plays, the passage being spoken is highlighted in blue, and the transcript scrolls to keep it in view. When word timings are available (cloud transcription, and Mac transcription with Apple's on-device speech), the word being spoken is highlighted too.

Click any line — or any word — to jump the audio to that moment. If you scroll the transcript yourself, it stops following for a few seconds so it doesn't pull you away.

The transcript is fully searchable — find who said what and when.

Downloading Transcripts

Export your transcript as:

• TXT — plain text format
• CSV — spreadsheet-compatible with speaker, timestamp, and text columns