Does an AI notetaker know who is speaking?

The details

• 01There are two different mechanisms and they are not equally reliable. A meeting app knows who is speaking because every participant has a tile and an active-speaker indicator. Voice-based separation guesses from the audio alone. When both exist, the name from the app wins.
• 02Speaker Tags, Reline's extension for Google Meet, Teams in the browser and Zoom in the browser, is rolling out and is not on for every account yet. Where it is on, it watches the active-speaker indicator and passes the name to the Reline app, which writes it onto the transcript line. Nothing joins the call, and the extension sends names, not audio.
• 03Voice-based separation of the far side is still in testing and is not on by default. Where it runs, it does well when voices differ and people take turns, and it is wrong more often with one room microphone, eight people and cross-talk. Treat its labels as an estimate and rename a speaker when you know who it was; the rename applies to every line.
• 04One thing is never a guess: your side versus theirs. Reline records your microphone and the system audio as separate sources, so what you said and what everyone else said are kept apart even when nothing else is known.
• 05The audio stays with the note. If a label looks wrong, press play on the line and hear who it was. That is faster than any attribution algorithm and it is the reason Reline keeps the recording.
• 06Labelling a speaker by name does not change the consent rules. Tell people you are recording and follow local law, whatever the transcript can or cannot infer.