Transcripts and speaker labels

Every recording becomes a transcript after processing, where each line has a speaker and a timestamp. Here’s what to expect, and how to fix the parts machines get wrong.

After you stop

The recording uploads and gets processed. A few minutes later (longer for long recordings), the note has the finished version:

• Speaker labels on every line — who said what.
• Clickable timestamps — click a line to jump the audio to that moment.
• Summaries — long calls are summarized in sections, so an hour-long call is skimmable.

How it knows who’s who

No bot joins your call. Banter uses the recording, meeting context, and the voices you’ve confirmed in earlier transcripts to suggest who said what.

The easiest way to get the best speaker attribution is to record a browser meeting with the Chrome extension. In Google Meet, Zoom, and Teams, the extension can read meeting metadata from the page, including participant names and speaking activity. That gives Banter real names to work with instead of relying on audio alone.

If you record without the extension, or record a native app or in-person conversation, Banter still separates speakers from the recording and suggests names where it can. Those suggestions are less informed because Banter cannot see the meeting page’s participant metadata.

Once you confirm who a voice is in the transcript, Banter can remember that voice and suggest the person in later recordings from your account. Confirming a name and remembering a usable voice are separate: the speaker picker shows whether the voice is remembered, still processing, or needs another attempt.

Fixing a name

Names Banter isn’t sure of are suggestions, shown with a question mark. If a voice labeled “Michael?” is really Priya, click the name in the transcript pane and pick Priya from the list, or type her name and choose Add “Priya” if she’s new. The choice applies to that whole voice, so every line it spoke shows Priya. To confirm a suggestion that’s right, click it and choose Confirm — or, for a name guessed from the meeting page, pick or add that person.

If the speaker picker says could not remember voice, use Retry remembering voice. If there is insufficient clear speech, confirm the person in a recording where they speak more. A correct name stays confirmed even if remembering the voice fails.

For recurring unnamed voices, open People → Needs a name to review and name selected conversations together. Listen and check the matches before saving; see People and organizations.

Older transcripts keep the speaker labels they were saved with; those can’t be changed.

Where transcripts show up

• In the note, as the recording’s transcript pane.
• In search — ask the assistant “when did Acme mention the security review?” and it finds the moment.
• To connected AI tools, which can read transcripts and quote exact wording with timestamps.