How Parley Notes learns who is speaking without a meeting bot or the cloud

Transcribe a meeting around a coffee table with a standard tool and you get a script full of "Speaker 1" and "Speaker 2". Half the value of the note is gone, because you no longer know who made the promise.

Meeting bots avoid the problem by cheating. Everyone joins from a separate video tile with a name on it, so the bot never has to tell voices apart. Put a phone on a table between three people and there are no tiles. There is one microphone and a room.

Teach it your voice once

Start with yours. Talk for fifteen seconds when the app asks, or open any note and tap This is me on one of your lines. On a note with a single voice the app simply asks, "Is this you?"

From then on your lines are marked as yours in every note, so the summary is written from your side of the table. "You promised the revised quote by Friday" is a different sentence from "Speaker 2 promised a quote".

Then the people you work with

When you rename a speaker in a note, or confirm or fix a name the app suggested, Parley Notes learns that voice too. Next week, when you sit down with the same client, their lines arrive already named. The people you meet most get recognised on their own over time.

It is deliberately careful. If there was too little of a voice to learn, if the lines mix two people, or if the voice sounds like yours or like someone already known, nothing is saved and the app says so. Recognition also works best close to the microphone. From across a noisy room a line may still read "Speaker 1" until you rename it, and we would rather say that plainly than promise what a phone cannot always do.

A voiceprint is biometric data and that is why it stays with you

A voiceprint is a short mathematical sketch of a voice, not a copy of the sound. It is still sensitive, so the design starts from one rule: every voiceprint, yours and those of the people your device knows, stays on that device and is never uploaded.

That holds even when the transcript itself is made in the cloud. The matching runs on your phone or computer. What leaves the device is a timeline of who spoke when, with the names you typed. No voiceprint goes with it.

You stay in charge of the list. Open Settings, then Your voice, then People this device knows. Forget one person or forget them all. Turn off Recognise my voice and your own voiceprint is deleted too. Voiceprints do not follow your account to a new phone.

The other people in the room

They did not choose the app, so the oldest habit still matters most: say you are taking notes before you start. Their voiceprints never reach our servers, and you can forget anyone the moment they ask.

Voice learning is in the Android and Windows apps.