How transcription works

Pulse transcribes meeting audio and uses that text to produce AI-enhanced notes. Transcription does not start on its own; it is triggered by you opening a note. This article explains when transcription starts and stops, which audio is captured, what you can do to improve accuracy, and how to get the finished text out.

#The four things that start transcription

Pulse starts capturing audio and transcribing only when you do one of the following. It never starts by itself.

1. Click the notification. Clicking the notification shown before a meeting starts opens the video meeting URL and, at the same time, opens that meeting's note and begins transcribing.
2. Open an event whose start time has passed, from the home screen. Opening a meeting from the home screen whose start time has already passed begins transcription at that moment.
3. Click [+ New note]. This opens a new note from the button at the top right. Use this for an unplanned discussion that is not in your calendar.
4. Leave the note for an upcoming event open. If you have the note for a meeting that is about to start already open, transcription begins automatically when the calendar start time arrives.

Put the other way round, a meeting whose note you have not opened is never transcribed. If a meeting has started but transcription does not appear to be running, check that you have that meeting's note (or a new note) open, and that the calendar start time has already passed.

#No recording bot joins

Pulse does not use a recording bot that joins the meeting as a participant. Processing runs on your own computer, taking sound from its system audio and microphone. No unfamiliar account appears in the participant list, and the meeting host does not need to allow a bot to join. Because of this, transcription happens only while the desktop app is running on your computer.

FigureScreenshot: the note screen during a meeting, with a green bar at the bottom showing audio being captured

The whole note screen with transcription running, the chat field at the bottom right and the icon for opening the transcript panel to its left.

#Telling whether transcription is running

While a note is open, a moving green waveform bar at the bottom of the screen means audio is being captured. If the bar is not moving, no audio is arriving, so check your input device using the steps in Transcription problems.

To the left of the “Ask anything” chat field at the bottom of the screen is a small icon shaped like an audio waveform. Clicking it shows or hides the transcript panel, where the converted text scrolls past as it is produced.

While another app is in front, an indicator floats at the right edge of the screen showing a meeting is in progress. Click it to return to the meeting's note, and drag the handle at its base to move it.

#What you can do in the transcript panel

Deleting by block is useful when unrelated small talk came up during the meeting, or when something was said that you do not want on the record. A deleted block is removed from the transcript.

#Reading the colours

Each block of the transcript changes colour and position according to where the sound came from.

• Grey bubble (aligned left): captured from system audio. Normally the speech of participants other than you.
• Green bubble (aligned right): captured from your microphone — that is, your own speech.

Speaker labels are not set up, this distinction is the only clue you have.

#What stops transcription

You can end transcription at any time with the stop button at the bottom of the screen. Beyond that, it stops automatically in the following situations.

• When the meeting is judged to have ended. That judgement uses the length of the transcript, whether you are still using a meeting app (Zoom, Microsoft Teams, Google Meet and so on), and, for a meeting tied to a calendar, the scheduled end time.
• When no new audio has been detected for fifteen minutes.
• When the computer is put to sleep.

#A caveat for meetings not tied to a calendar

For a meeting linked to a calendar event, the scheduled end time is one of the signals used to decide it has ended. An unplanned discussion started from [+ New note] has no scheduled end time, so the only clue is whether the meeting app has stopped using the microphone.

As a result, if you carry on talking in the same room after hanging up, the audio never breaks, the fifteen-minute silence timer never fires, and transcription can keep going. For meetings not in your calendar, pressing the stop button when you finish is the surest approach.

#When automatic stopping does not work on macOS

On macOS, deciding automatically that a meeting has ended requires administrator rights on that Mac. On company-issued machines without administrator rights, automatic stopping does not work, so press the stop button yourself when the meeting ends.

#Back-to-back meetings

Going straight from one meeting into another can leave the meeting app holding the microphone through the gap. In that state the end of the first meeting cannot be detected, and the two meetings can end up joined into a single transcript. If you want a separate note for each meeting, press the stop button in between.

#Resuming after it has ended

Even after a meeting has ended and the note has been generated, you can resume transcription in the same note. Open the transcript panel (the waveform icon to the left of the chat field) and click “Resume”, and new audio is appended to the existing transcript. Regenerate the note afterwards and the later material is reflected in the summary too.

#Which audio is captured

Pulse captures your computer's system audio and microphone input. System audio is the sound your computer is playing, which normally includes the speech of participants other than you. Microphone input is your own speech. Any call whose audio plays through the computer can be transcribed, including VoIP and internet calling apps. Those calls are not linked to a calendar, so start from [+ New note] and check that your audio device settings match the input and output the meeting is using.

Audio cannot be separated per app. What is captured is the system audio as a whole. If you play music or a video during a meeting, that sound is included in the transcript even if you have sent it to a different output device.

For the purpose of transcription, audio passes directly from your microphone and system audio to the transcription processor UNCHAIN, Inc. engages. There is no mechanism that records or stores the call's audio or video itself, so a meeting's audio file cannot be retrieved afterwards. For how long transcript text is kept, see Automatic transcript deletion and Our security standards.

Pulse is also designed around transcribing a meeting as it happens. It cannot load and transcribe an audio file such as an MP3 recorded in advance.

#Telling speakers apart

Speaker labels puts participants' display names in the transcript instead of “me” and “them”. Which meeting tools are supported, what setup is needed, and what can be identified are covered in Speaker labels.

Where speaker labels are unavailable or not set up, the transcript shows only “me” and “them”. “Me” corresponds to microphone input and “them” to system audio. Even so, who said what is often inferred from the surrounding context and reflected in the note; if a name is wrong you can correct it directly in the note or tell chat to fix it.

#Improving accuracy

Transcription accuracy depends heavily on the quality of the incoming sound and on what vocabulary you have registered. Check these three things.

1. Register proper nouns. Company names, product names, in-house terminology and colleagues' names go into your personal jargon list under language settings in Settings. If an administrator has set workspace-level jargon, that applies too. The steps are in Customising transcription.
2. Match your audio device settings. The input and output devices your meeting app uses must match the device Pulse is taking sound from. Also check the microphone input level is not too low.
3. Use a headset where it helps. This suppresses background noise and sound from the speakers looping back into the microphone. It makes the most difference in shared office space and in meetings where several people join from the same room.

#Exporting the transcript

Transcript text can be copied from the panel and pasted into other tools.

1. Open the meeting's note and click the waveform icon to the left of the chat field at the bottom of the screen to open the transcript panel.
2. To copy the whole thing, click the copy icon at the top right of the panel.
3. If you only need part of it, click the block you want and then the copy icon that appears.
4. Unwanted blocks can be removed from the transcript using the delete icon that appears in the same way.

To pass a whole note to another tool, use the steps in Exporting notes.

#If something is not working

• Transcription does not start: check that a note is open and that the calendar start time has passed. If it still does not start, check your audio device settings and restart the app. Detailed steps are in Transcription problems.
• Two meetings ended up in one transcript: this is likely because the meeting app kept hold of the microphone. Press the stop button between meetings next time.
• It does not stop when the meeting ends: automatic stopping does not work on macOS without administrator rights. End it manually with the stop button.

If only the other party's voice is missing, system audio may not be being captured. Where to check depends on your operating system.