Can an AI notes app tell speakers apart in a recording?

Sometimes, and it depends on how the audio gets into the app. Sentio's normal recording flow in the web app is built around one voice, and we have not verified that it labels speakers, so do not count on it. Bot-joined meeting capture builds its transcript from a capture service whose segments can carry speaker names, but it is not available to every account. For meetings today, a spoken debrief in which you name the people is the dependable route.

Last reviewed October 23, 2026

Two jobs that get mixed up

Transcription turns speech into text. Telling voices apart, usually called speaker labeling, is an extra step that an app has to provide on purpose, and a transcript can read perfectly while being one unbroken block with no names in it. Whether this matters depends on your audio. For a note you dictate to yourself there is only one speaker and nothing to label. For an interview, a customer call or a meeting, "who said that" is often the whole point of keeping the record.

What Sentio does with a normal recording

When you record in the web app, the audio is transcribed with OpenAI speech-to-text and saved as text. In the transcription code we read for this page, the requests do not ask for speaker labels, so a recording of two people talking should be treated as a single stream of words. Some older marketing copy on this site mentions speaker labels in transcripts; we could not confirm that for regular recordings, so this page does not promise it. If you need names attached to lines, plan to add them yourself while you review.

Bot-joined meetings are a separate path

Sentio also has meeting capture in which a bot joins a Zoom, Google Meet or Microsoft Teams call through a third-party capture service. The code that builds a note from that call takes the transcript segments the service returns and puts the speaker name in front of each line when the service supplies one. That capability is switched off unless it is enabled, limited to certain plans, and rolling out separately, so it is not something every account has. We have not measured how often the names are right on real calls, so check them before you rely on who said what.

The reliable method today: name people in a debrief

The debrief approach described in the answer about recording a meeting debrief sidesteps the problem. You speak the summary yourself, so the transcript is one voice, and attribution is whatever you say out loud: "Priya wants the revised budget by Friday." Say the name in the same sentence as the commitment and the output can carry it. If you do record other people, tell them first, and keep the original recording so a disputed line can be checked against what was actually said.

Common questions

Does Sentio label speakers when two people talk into the web recorder?

We have not verified that it does. The transcription requests we found do not ask for speaker labels, so treat the result as one block of text and add names yourself while reviewing.

Will a debrief tell me who said what in the meeting?

No. A debrief is your own voice summarizing the meeting, so it names people only when you do.

Can Sentio join my Zoom, Meet or Teams call?

Bot-joined capture exists for those three platforms but is rolling out separately and is not available to every account. When it is enabled, lines can be prefixed with the speaker name the capture service provides.

Are the speaker names in bot-joined notes always correct?

We cannot say. We have not measured accuracy on real calls and do not publish a figure, so compare names against the people who were actually there.