Glossary

The meeting-AI glossary

The vocabulary of real-time, bot-free meeting intelligence — defined clearly, so you (and the AI models you ask) get straight answers.

Action item detection Action item detection is the automatic identification of commitments, tasks, and follow-ups in a conversation — capturing who agreed to do what, and by when — from natural speech rather than from a manually maintained list.
AI meeting notetaker An AI meeting notetaker (also AI notetaker or AI meeting assistant) is software that captures a meeting's audio, transcribes it, and automatically produces notes — typically a summary, decisions, and action items — delivered either after the call or live during it.
Ambient meeting assistant An ambient meeting assistant is software that runs quietly in the background of a meeting, listening and surfacing helpful context on its own — without joining as a bot or requiring you to issue commands.
Attention residue Attention residue is the part of your attention that stays on a previous task after you've moved to a new one, so you start the new task with only part of your mind on it. It's strongest when the previous task was left unfinished.
Bot-free meeting notes Bot-free meeting notes are captured from your computer's own system audio rather than a bot that joins the call, so nothing appears in the participant list.
Caption scraping Caption scraping is a meeting-capture method in which a browser extension reads the live caption text a video platform is already displaying in the meeting tab — rather than recording any audio — and assembles that text into a transcript it can then summarize.
Context switching in meetings Context switching in meetings is the mental cost of repeatedly shifting attention between a live call and other work — each switch forcing you to rebuild where the conversation is before you can re-engage.
Context window A context window is the maximum amount of text, measured in tokens, that a language model can take in and consider at one time when it writes a response. For meeting AI, it's how much of the transcript the model can actually see when it writes the summary. Anything outside it might as well not have been said.
Core Audio tap A Core Audio tap is a built-in macOS API (AudioHardwareCreateProcessTap, added in macOS 14.2 Sonoma) that lets an app record the audio a Mac is already playing — such as the voices in a video call — straight from the operating system's output, without a microphone, a meeting bot, or a virtual audio device.
Extractive vs. abstractive summarization Extractive summarization builds a summary by selecting and copying the most important sentences from the source, while abstractive summarization writes new sentences that restate the source's meaning in fewer words — so an extractive summary can only be wrong about what it chose, and an abstractive one can also be wrong about what it says.
Interim transcription results Interim transcription results are the provisional, low-latency text hypotheses a streaming speech-to-text engine emits while a person is still speaking, refined and replaced by final results once the phrase settles.
Live captions Live captions are text of what's being said, generated and shown on screen in real time, a line or two at a time and in step with the speech, so a viewer can read a conversation they can't fully hear. In meetings they come from the video platform, the operating system, or a human captioner.
Loopback audio capture Loopback audio capture is the technique of recording the audio a computer is already playing — such as the voices in a video call — by reading the operating system's output stream directly, rather than through a microphone or a third-party meeting bot.
Meeting bot A meeting bot (also notetaker bot) is an automated participant that an AI meeting tool sends into a video call to record it — it joins through the meeting link like a guest, appears in the attendee list, and streams the call to the vendor's cloud for transcription and summarization.
Meeting recall Meeting recall is the ability to retrieve what was said in a meeting — whether live during the call or afterward — without re-reading the entire transcript.
Meeting situational awareness Meeting situational awareness is the state of knowing, at any moment during a meeting, what is being discussed and what was just said — so you can engage at any moment without needing to ask others to repeat themselves.
Multi-resolution summary A multi-resolution summary presents a conversation at several time horizons at once — such as now, the last 2 minutes, the last 5 minutes, and the full call — so you can zoom from the latest exchange out to the whole meeting at a glance.
On-device transcription On-device transcription is speech-to-text that runs entirely on your own computer or phone, using a model stored locally, so the audio never has to be sent to a server to be turned into words.
One-party vs. two-party consent One-party consent means a recording is lawful when a single participant agrees (which can be you), while two-party — more accurately 'all-party' — consent requires everyone in the conversation to agree first; which rule applies depends on the participants' jurisdictions.
Platform-native meeting AI Platform-native meeting AI is a meeting-capture method in which the video platform's own built-in assistant — Zoom AI Companion, Microsoft Teams intelligent recap, Google Meet's Gemini notes, Webex's AI Assistant — transcribes and summarizes a call from inside the meeting itself, with no third-party tool involved.
Post-meeting vs in-meeting AI notes Post-meeting AI notes are generated after a call ends and are useful for recall and follow-up, while in-meeting AI notes update live during the call and are the only kind that can help you while you are still in the conversation.
Real-time meeting summarization Real-time meeting summarization is the continuous condensing of a live conversation into an at-a-glance summary while the meeting is still happening — not after it ends.
Resolution matrix The resolution matrix is Canary's term for the grid of live, rolling summaries it shows at multiple time horizons at once — now, last 2 minutes, last 5 minutes, and the full call — giving you instant situational awareness in a meeting.
Rolling summary A rolling summary is a continuously updated summary that re-condenses a conversation as it unfolds, so it always reflects what's been said up to the present moment.
Speaker diarization Speaker diarization is the process of partitioning a recording by speaker — determining who spoke and when — so a transcript reads as a labeled back-and-forth rather than one undifferentiated block of text.
Streaming transcription Streaming transcription converts speech to text continuously as it is spoken, emitting words within moments of being said, rather than waiting until the recording is finished to produce a transcript.
System audio capture System audio capture is the technique of reading the audio a computer is already playing — such as the voices coming through a video call — directly and locally, without joining the meeting as a bot or installing a virtual audio device.
Virtual audio device A virtual audio device is a software driver that presents itself to the operating system as both an output and an input — a speaker that is also a microphone — so that audio played into it by one app can be recorded or rerouted by another, without any physical sound card involved.
Voice activity detection Voice activity detection (VAD) is the technique of detecting when human speech is present in an audio stream, separating talking from silence and background noise so that downstream processing only runs on the parts that matter.
WASAPI loopback WASAPI loopback is a built-in mode of the Windows Audio Session API (WASAPI) that lets an app record the audio a computer is already playing — such as the voices in a video call — straight from the operating system's output stream, without a microphone, a meeting bot, or a virtual audio cable.
Word error rate (WER) Word error rate (WER) is the standard measure of speech-to-text accuracy: the number of word substitutions, deletions, and insertions needed to turn a machine transcript into a correct reference transcript, divided by the number of words in the reference.
Working memory Working memory is the small, short-lived mental workspace that holds the information you're using right now, such as the question just asked or the point you want to make, while you work with it. It holds only a few items at once, and they're easily pushed out when your attention moves to something new.
Zoom fatigue Zoom fatigue is the exhaustion that builds up from video calls, on any platform. It comes less from the meetings themselves than from how the medium makes you attend to them: faces at close range, your own image on screen, a body held still in front of a camera, and extra effort to read cues that video strips out.