Skip to content

Record a meeting with the note taker

The note taker is an element of the AI Toolbox that transcribes a recorded meeting and says who said what, with the time. You record the meeting in your browser or upload a recording, give the speakers names, and pass the text on: to an element that turns it into minutes or an action list, to a chat with an assistant, to a project or to a knowledge base. This page explains how that works, how long OS keeps the recording and who gets to see what.

In the toolbox you recognise a note taker by the word Note taker beside its title. An element that transcribes recordings without recognising speakers works the same way: you then get the text with the time of each part, without speakers and without names.

Record or upload a recording

A recording may last up to 2 hours. The note taker records with one microphone: your computer's or your phone's, or the headset you use.

  1. Open the note taker in the AI Toolbox.
  2. Tell the people in the meeting that you are recording it. OS reminds you: Tell the people in this meeting that you are recording it.
  3. Choose Record and click to start the recording. You can pause and resume it. When you are done, choose Complete recording. If you already have a recording, choose Upload file and drag the file there or click to browse.
  4. If the form asks How many people are speaking? (optional), fill in roughly how many people spoke. It helps the model tell the voices apart. You only see the question for a model that can use it.
  5. Choose Start. The recording is sent; while that lasts you see a bar and Keep this tab open until sending is done. Stop cancels the sending.

While you record, OS warns you when it hears no sound (No audio detected — check your microphone.) and a few minutes before the recording stops at its limit. If you close the tab or your browser crashes while recording, the part already recorded is not lost: when you come back in the same browser, the note taker offers the recording with Send it or Discard. That recording exists only in that browser.

When OS does not accept a recording

  • This recording is longer than 2 hours. Cut it into parts and send them separately.
  • This file cannot be read here. Convert it to MP3 or M4A and try again.
  • This model takes recordings of up to a number of minutes: this note taker's model takes shorter recordings. Choose a shorter recording or ask your administrator for another model.
  • Sending failed. The recording is still ready; choose Send again.

Your browser first makes a large file smaller before it is sent (Making the recording smaller…); for a long recording that can take a while. If you see This recording may be counted as longer than it lasts., the file does not state its own length. The recording is transcribed all the same, but may weigh more heavily on your organisation's monthly budget.

While the recording is transcribed

After sending, the recording appears among your results with Being transcribed, and how long it has taken so far or how many parts are done. You do not have to wait: you can close the page and carry on. When the text is ready, you get the notification Your transcript is ready. If it fails, you get Your recording was not transcribed, by default by mail too, so you can try again while the recording is still kept. Whether you get notifications by mail, you choose in your notification settings. The notification takes you to the result.

A two-hour meeting is usually transcribed within a few minutes. If it takes exceptionally long, OS stops after six hours and you are notified that it did not succeed.

When a transcription fails you see Transcription failed with the reason. If the recording is still kept, Retry transcription tries again without recording again. If the recording is no longer kept, record again or upload the file again.

Name the speakers

The result shows the text line by line, with who spoke and when. The speakers are first called Speaker 1, Speaker 2 and so on, in the order they first speak. That number belongs to this recording only: Speaker 2 in one recording has nothing to do with Speaker 2 in the next.

  1. Open Speakers. For a result without names the list is already open.
  2. Fill in the names under Name of speaker 1 and the ones after it. A name is saved as soon as you leave the field, and the text shows it straight away.
  3. If you do not know who a speaker is, choose Play from and the time: you hear the first thing that speaker said. This works while the recording is kept.

If the model heard one person as two speakers, give them the same name. The two are then merged: the text shows that name for both, and the list names it once, with Merged with and the name. To remove a name, empty the field.

Names change nothing in what was said: OS keeps the text as it was transcribed and puts the names beside it when you read it. A name that happens to appear in someone's sentence therefore stays as it is. If you changed the names in another tab, OS says The names were changed elsewhere meanwhile; you now see the latest.

Have names suggested

If the people in the meeting call each other by name, OS can suggest the names. Choose Suggest names. Suggestion: with a name then appears beside a speaker; choose Accept or Decline. A suggestion is only saved once you accept it.

A suggestion comes from what is said, never from the voice. OS recognises no one by their voice and remembers no voices or names from one recording to the next. Every suggestion is checked: OS keeps it only when the name appears word for word in what that speaker, or the speaker before or after, said, for example Thanks, Jan, then the budget. A name the model makes up is dropped that way. A suggestion can still be wrong, since whoever says a name is not always talking to that person. So check it before you accept it. If OS finds nothing, it says No names found in the text.

A suggestion is a request to your organisation's helper model. If your organisation shows costs, how many tokens it used appears beneath it; otherwise you see no usage.

Read, enlarge and download

Enlarge opens the result across the whole screen, to read a long text calmly and fill in the names. Under Download you find Download recording, while the recording is kept, and Download transcription, the text with names and times as a text file.

Pass the text on

Choose Pass on to do something with the text. You can as soon as the text is there:

  • Minutes and Action list, the two elements your organisation gets by default, or other next steps the note taker's maker chose. Minutes writes a report with the attendees, the points that came up, what was decided and the actions. Action list makes a table with who, what and when for each action, with the line of the text the action comes from. Both write in the meeting's language and add nothing that was not said.
  • Another element… opens Pass on to an element: choose an element that takes a document and choose Pass on. The element opens with the text already filled in.
  • Chat with an assistant… opens Start a chat from this output: choose an Assistant, write a Message (optional) if you like, for example Make a summary of this., and choose Start chat. You only see assistants that take files.
  • Add to a project… puts the text among the files of a project you may add to. Everyone in the project can read it.
  • Add to a knowledge base… puts the text as a document in a knowledge base you may add to. Whoever uses the knowledge base can find it.

What you pass on is a copy of the text as it is at that moment, with the names. Names you change afterwards do not follow into copies you already passed on. So name the speakers first. From then on a copy follows the rules of the place it is in: a project or knowledge base decides who reads it, and a copy in a chat goes with the chat.

How long the recording is kept

OS keeps the recording for a while, so you can listen back to who was speaking. The result says The recording is kept until and the date. How long that is, your organisation chooses: 30 days by default (Organisation settings). After that OS deletes the recording by itself. The text and the names stay; you can then no longer listen back or download the recording.

To get rid of the recording sooner, choose Delete now beside that date, or Delete recording under More, and confirm with Delete recording. The text and the names stay.

The text itself is kept like every result in the toolbox, until the period your organisation set for runs and content.

Delete the result

Choose Delete result under More to delete everything of this recording: the text, the names and the recording. If the recording is still being transcribed, OS stops that. Copies you already passed on to an element, a chat, a project or a knowledge base stay where they are.

Who sees it

Only you. Your recordings, the text and the names are yours, like every result in the toolbox. Your organisation's administrators do not see them. Someone else reads the text only when you pass a copy on to a project or a knowledge base.

What the model's provider hears

The recording goes to the provider of the transcription model the note taker uses. Which model that is and in which region it runs follow your organisation's choices. A recording cannot be pseudonymised: the voices and every name said in the meeting reach the provider as they were spoken, even where your organisation hides personal data elsewhere. So do not discuss anything in a recorded meeting that must not reach that provider.

Your organisation is responsible for what happens to the recordings. You, as the person recording, tell the others in the meeting that you are recording.

Make a note taker

If you may change the toolbox, you make a note taker as an element that transcribes recordings. In the form you choose the Transcription model, and if you like the Language and Guidance, such as names and spellings that come up often.

  • Recognise speakers turns the element into a note taker: the result says who said what, and the element is then called Note taker. You only see the switch for a model that can recognise speakers.
  • Next steps are up to 5 elements that take a file; they appear under Pass on on the result. A new note taker gets Minutes and Action list, where your organisation has them. You can remove them or choose others.

OS makes Minutes and Action list once for every organisation, as soon as it has a helper model. They are ordinary elements: you can change, rename or delete them, and OS does not make a deleted one again.

If you give the note taker to an assistant, the assistant transcribes a recording with it within its own answer, without recognising speakers. That suits a short recording; you record a meeting with the note taker itself in the toolbox.