Transcription labels each distinct voice speaker_0, speaker_1 and so on, numbered per recording — so speaker_1 in one session is not necessarily the same person as speaker_1 in the next. Speaker detection then tries to match those labels to the people and characters it already knows from your campaign. Where it is confident, the session carries on. Where it is not, the session is flagged for review.
When review is asked for
A session needs speaker review when any detected speaker is unconfirmed, has no display name, has a role of unknown, or when the automatic matching isn't confident who the GM is. The session page shows Name voices with the note that highlights and recaps use speaker labels until you do; you also get an email and, if configured, a Discord post.
Transcript-only sources (pasted or uploaded documents, past-session notes) skip detection. YouTube captions arrive as one speaker, so they always need a pass.
Where to do it
- Wrap-up page (the Review link on the session, or Overview → Catch up). The first section, Speakers — who is who, lists each voice with a sample quote. Type the character or real name and the speaker is confirmed.
- Full speaker roster (
/campaigns/<your-campaign>/speakers) shows every label the campaign has met, grouped under the recording it was heard in. Each card has a Display name and a Role of GM, Player or Unknown; edits save when you click away. You can add a speaker by hand too — a Label (required, e.g.speaker_0), the Recording it belongs to (or Every recording for a campaign-wide default), a display name and a role.
What confirming does
A confirmed speaker carries forward: the next session's detection starts from what you have already named, and processing never overwrites a confirmed row. A name you confirm against one recording stays attached to that recording, because the numbering is per recording. Recaps written in the story form use character names and remove players' real names.