Skip to content
Memory Notch
Memory NotchSpeaker RecognitionMAC · IPHONE

SPEAKER RECOGNITION

Name someone once.
Memory Notch knows them next time.

Give a speaker a name in any note. Memory Notch learns that person’s voice and fills in their name in your other recordings, past and future. It works on your device, and nothing is uploaded.

ILLUSTRATION · NOT A SCREENSHOT

Short answer: Speaker Recognition is off until you turn it on in the Personalize tab. Once it is on, naming a speaker in one note teaches the app that voice. It then fills in the name in other recordings where it hears the same voice. It only uses names you have given.

Turn it on

Speaker Recognition is off by default. Open the Personalize tab and switch it on.

Personalize is the tab that used to be called Vocabulary. Your vocabulary list is still there.

Name someone once

Speakers start out numbered: Speaker 1, Speaker 2. Give one of them a name in any note. Memory Notch learns that person’s voice, and from then on it fills in their name in other recordings, past and future.

New recordings are named when the recording stops. While you are recording, speakers are still numbered.

It works for everyone in your recordings, not only you.

See what the app filled in

A name the app filled in is marked as recognised until you confirm it. Each one comes with Confirm, Not them and Undo, so you can keep a name or take it back.

When a match is less certain, the app does not apply it. It offers the name as a suggestion and leaves the choice to you.

Scan the recordings you already have

Scan my library goes through your existing recordings once and fills in the names of the people you have named.

Who is this?

Some voices turn up again and again before you have named them. Memory Notch groups unnamed voices it has heard in more than one note and asks: Who is this?

  1. Play a short clip of the voice.
  2. Type a name once.
  3. The name is filled in wherever that voice is heard.

The app does not work out who anyone is. It only notices that the same unnamed voice has come up in more than one note. The name always comes from you.

Where your voice data lives

Speaker Recognition works by voice, and all of it happens on your device. It uses a small on-device voice model (7.6 MB) and plain distance maths to tell whether two voices match. There is no cloud service and no account, and nothing is uploaded.

Voice data is stored on your device, outside your notes. The privacy policy has the full wording.

Delete it whenever you like

Delete all voice data removes it with one button. You can also delete the voice data for one person.

What it doesn’t do

  • No names during a recording. Names appear when the recording stops.
  • No names you haven’t given. It fills in the names of people you have named, and nobody else.
  • No certainty. A recognised name is the app’s match, not a verified identity. That is why it stays marked until you confirm it.
  • Not on Windows or Android yet. Those apps label speakers with numbers.

Questions

Does Speaker Recognition send my voice anywhere?

No. It works entirely on your device, with a small on-device voice model (7.6 MB) and plain distance maths. There is no cloud service and no account, and nothing is uploaded.

Is Speaker Recognition on by default?

No. It is off until you switch it on in the Personalize tab.

Does it name speakers while I am recording?

No. While you record, speakers are numbered. Names are filled in when the recording stops.

Can it identify someone I have never named?

No. Memory Notch only fills in names you have given. Who is this? groups unnamed voices it has heard in more than one note, and the name still comes from you.

How do I delete the voice data?

Delete all voice data removes it with one button. You can also delete the voice data for one person.

Download for Mac · iPhone and iPad · All features · Privacy policy