How Noats tells speakers apart without learning your voice

A transcript with three people in it is only useful if you can tell who said what. The decision that the customer is not renewing means something different depending on whether the customer said it or your colleague did.

Most tools solve this by learning what each person sounds like and matching them across every call. That is a biometric record, and it lives on their servers. Noats takes a different route.

What happens during a call

Everything that comes through your Mac's audio, which is every voice on the other end of a call, is split into segments. Each segment is turned into a compact description of the voice, and segments that describe the same voice are grouped together. That grouping happens on your machine, while the meeting is still running.

The result is a set of unnamed speakers. You give each one a name when you like, during the call or after it, and the transcript updates. If two people share a microphone at the far end of a call, they still come apart, because it is the voice that is being grouped, not the channel.

Your side of the call is always you

Whatever your Mac's microphone hears is attributed to you, so your own words are labelled correctly from the first second, with nothing to set up.

Why the names do not follow you into the next meeting

The names you assign belong to that meeting. Tomorrow's call starts fresh, and so does a recording you resume after a break. There is no enrolment step and no profile of anyone's voice on your disk.

We could build that profile. We chose not to, because a file that can identify a person by the sound of their voice should not exist on a laptop by default. Naming a speaker takes a few seconds, and it means the people you talk to are never turned into data.

A voiceprint is a key to a person. Noats does not cut keys.

What this looks like in the notes

• The live transcript shows speaker labels as they are identified, and you can rename them in place.
• The written-up note uses the names you chose, so the summary reads as a conversation between people, not between Speaker 1 and Speaker 2.
• Renaming after the fact changes every line that speaker said.

Speaker separation runs on Apple silicon alongside transcription, with no network involved. The privacy page has the complete picture of what stays on your Mac.