Videos with several speakers: who said what

What you'll see

When two or more people talk in a video (interviews, conversations, podcasts, co-hosted shows), we first work out who is speaking when from the audio, then write the chapters. So chapter points and summaries say "Jane says…" or "the guest shares her own…": the host's experiences are no longer attributed to the guest, and the guest's views are not attributed to the host.

Videos with a single speaker (tutorials, commentary, news) are not affected at all.

Where the names come from

• Introductions, how people address each other and how the host introduces guests in the video, plus the hosts and guests listed in the description. Host names found in other public analyses of the same channel are used only to fill in full names.
• Anyone we cannot name gets a role label such as "Host A" or "Guest B". English shows with several hosts often end up with labels only.
• Occasionally a name is matched to the wrong voice. If you spot it, edit that section directly, or use "Regenerate" and say who is who in your note.

Limits

• People with similar voices, or panels with a lot of cross-talk, can still be mixed up.
• At most 8 speakers per video; videos longer than 6 hours skip this step.
• Quoted material ("a study found…") is still written as the speaker's own view.

Credits and privacy

• No extra credits; it is part of the analysis.
• To tell speakers apart, the video's audio is sent to a speech-processing service we run ourselves. Audio passes through only during the analysis, is not kept, and no voiceprints are ever created.