GlossaryThe words behind talking instead of typing.

22 definitions, in plain English. Generic terms get generic definitions — where WisprBee does something specific, that's noted separately so you can tell the two apart.

Voice keyboard

Also called: AI voice keyboard

A voice keyboard is software that produces the finished text you would have typed, from speech, formatted for wherever the text is going. It differs from dictation in scope: dictation converts speech to words, while a voice keyboard also does the punctuation, cleanup and formatting work that stands between a transcript and something you can send.

In WisprBee: WisprBee is a voice keyboard for macOS. It runs at the OS level rather than inside one app, so the same key works in any text field on the machine.

Dictation

Also called: Speech-to-text, STT, Voice typing

Dictation is the conversion of spoken audio into written text. Classic dictation output is a transcript: it reflects what was said, including filler words, false starts and no formatting, and normally needs editing before it can be used.

App-aware dictation

Also called: Context-aware dictation, Destination-aware dictation

App-aware dictation is dictation that changes its output based on which application will receive the text. The same spoken sentence becomes a short chat message in a messaging app and a structured email in a mail client, without the user selecting a mode.

In WisprBee: WisprBee's Context Engine detects the frontmost application the instant you press your key — and, in a browser, the website you're on — and picks one of eight destination profiles from it.

Context Engine

The Context Engine is the WisprBee component that identifies where your text is going and shapes it accordingly. It reads the frontmost macOS application (and the current website, in a browser) at the moment you press your push-to-talk key, then applies the matching destination profile to the text before pasting.

In WisprBee: It requires no configuration and offers no mode picker, which is the point: the cost of choosing a tone is exactly the cost the feature exists to remove.

Destination profile

A destination profile is a set of formatting rules for one kind of writing target — chat, email, code prompt, document, ticket, social post. Each profile decides structure, tone, length and conventions like whether to include a greeting or a sign-off.

In WisprBee: WisprBee ships eight profiles. Chat gets tight, friendly text with no sign-off; email gets a greeting, paragraphs and a closing; coding and AI tools get file names @-tagged and identifiers in backticks; docs get headings and bullets; issue trackers get a title, description and acceptance criteria; social gets a hook and short spaced paragraphs.

AI notetaker

Also called: AI note taker, AI meeting notes, Meeting assistant

An AI notetaker is software that captures a conversation, transcribes it, and turns the transcript into usable notes. The category splits on how the audio is captured: some tools join the meeting as a participant, others record from the user's own device.

In WisprBee: WisprBee's Notetaker captures from the device — your microphone and your Mac's system audio together — so it hears every voice on the call without joining it. It also handles long-form voice notes when there is no meeting.

Bot-free capture

Also called: Botless notetaker, No-bot meeting recording

Bot-free capture means recording a meeting from a participant's own device rather than by adding an automated attendee to the call. The practical differences are that nothing appears in the participant list, no one has to admit or approve a bot, and the tool works on any platform because it never integrates with one.

In WisprBee: This is how the WisprBee Notetaker works. It captures the audio your Mac is already playing, so Zoom, Meet, Teams and a phone on speaker are all the same to it.

System audio capture

Also called: Loopback audio, Internal audio recording

System audio capture records the sound a computer is playing — the other people on a call, a video, anything coming out of the speakers — as distinct from microphone capture, which records only the sound in the room. Capturing both together is what makes a recording of a call contain both sides of it.

In WisprBee: The Notetaker captures microphone and system audio together. Dictation, by contrast, only ever uses the microphone.

All-party consent

Also called: Two-party consent, Recording consent

All-party consent is the legal requirement that everyone in a conversation agrees before it may be recorded. Twelve US states apply it — California, Connecticut, Delaware, Florida, Illinois, Maryland, Massachusetts, Montana, New Hampshire, Oregon, Pennsylvania and Washington — while most others require only one party to consent. Where participants are in different states, the strictest applicable rule governs, and an AI notetaker counts as recording.

In WisprBee: Obtaining consent is the user's responsibility, not the software's. Asking at the top of the call is normally sufficient; California has additionally required disclosure of virtual notetakers since January 2025. This is information, not legal advice.

Push-to-talk

Also called: PTT, Hold-to-talk

Push-to-talk is an input pattern where audio is captured only while a key is physically held down, and capture ends when it is released. Compared with a toggle, it removes the failure mode of a microphone left recording, and it makes the end of the utterance explicit rather than guessed from silence.

In WisprBee: WisprBee's dictation key is push-to-talk and can be remapped to Fn/Globe, a Command, Option, Control or Shift key, or F13–F19. Command mode gets its own separate key.

Raw transcript

A raw transcript is the unedited text output of a speech-to-text model, before any cleanup or formatting is applied. It is the record of what was actually said, which makes it the thing worth keeping when later processing steps are lossy.

In WisprBee: WisprBee preserves the raw transcript of every dictation before polishing, at every cleanup level. That is what makes re-transcribe, undo and the diff view possible, and it means a bad polish can never cost you your words.

Auto-Polish

Auto-Polish is WisprBee's cleanup pass: it removes filler words, fixes grammar, punctuation and capitalisation, and resolves spoken self-corrections — so “Tuesday — no, make it Thursday” becomes Thursday, not both.

In WisprBee: It has four levels, from an exact transcript with no edits through to a full rewrite. The raw transcript survives at every level, so turning the polish up is never destructive.

Re-transcribe

To re-transcribe is to re-run the processing of a saved dictation without speaking it again, producing a new output from the stored original. It is only possible if the raw input was kept.

In WisprBee: Any item in your WisprBee history can be re-transcribed — useful for re-running the polish at a different level, or after you've taught the dictionary a term it got wrong the first time.

Command mode

Also called: Voice editing, Edit by voice

Command mode is editing existing text by voice: you select text, hold a dedicated key, and speak the version you want, and the spoken version replaces the selection. It differs from dictation in that the output overwrites a selection rather than being inserted at the cursor.

In WisprBee: It works in any editable field on macOS, uses its own hotkey so it can't collide with dictation, and returns a complete rewrite of what you said rather than an abbreviation of it.

Personal dictionary

Also called: Custom vocabulary, Word list

A personal dictionary is a user-supplied list of names, product terms, acronyms and unusual spellings that a speech system should get right. It exists because proper nouns and jargon are where general-purpose transcription is weakest — the model has no way to know how you spell a colleague's name.

In WisprBee: WisprBee's dictionary corrects by sound rather than by spelling, so a term is fixed even when the transcript came back as a different but similar-sounding word. Basic holds 50 terms; Pro and Pro Plus are unlimited.

Phonetic matching

Also called: Fuzzy matching

Phonetic matching compares words by how they sound rather than how they are spelled, so that similar-sounding strings can be treated as the same word. It is what makes correcting a mis-heard name possible at all: find-and-replace cannot fix “Geoffrey” from an entry for “Jeffrey”, because the string it would need to search for is the one it never saw.

In WisprBee: This is how the WisprBee Personal Dictionary applies your terms, rather than by literal substitution.

Adaptive dictionary

An adaptive dictionary is a personal dictionary that populates itself from the user's corrections, rather than requiring entries to be added by hand. Every time you fix a word, the fix is learned.

In WisprBee: The Adaptive Dictionary is on Pro and Pro Plus. It is the one capability the free plan genuinely does not have, as opposed to having a smaller quota of.

Snippet

Also called: Text expansion, Voice shortcut

A snippet maps a short trigger to a longer block of text, so the block can be inserted without typing it out. In a voice tool the trigger is a spoken cue rather than an abbreviation.

In WisprBee: Saying “my scheduling link” pastes your calendar URL. Basic includes 15 snippets; Pro and Pro Plus are unlimited.

Dictation history

Dictation history is the stored record of past dictations, normally searchable and re-usable. Its value is that speech is otherwise a one-shot: without a history, a dictation you didn't paste correctly is simply gone.

In WisprBee: Every WisprBee dictation is saved locally, searchable, and tagged with the app it went to and the style used. Basic keeps 30 days; Pro and Pro Plus keep it indefinitely. It lives on your Mac unless you turn on cloud sync.

Accessibility permission (macOS)

Accessibility permission is the macOS privacy control that allows an application to observe and control other applications — including inserting text into another app's fields. macOS requires it for any tool that types on your behalf outside its own window, which is why every system-wide text utility has to ask for it.

In WisprBee: WisprBee needs it to paste into other apps, and requests it during setup along with microphone access. Without it, dictation would only work inside WisprBee itself.

Notarization (macOS)

Notarization is Apple's automated security check of a macOS application, after which Gatekeeper will open the app without warning the user. An app that has not completed it may be reported as “damaged” or unverified, which reflects the missing check rather than a problem with the file.

In WisprBee: WisprBee is signed with Apple Developer ID and notarized, and the ticket is stapled to both the app and the disk image — so it opens without a warning even on a Mac that is offline.

Merchant of Record

Also called: MoR

A Merchant of Record is the legal seller in a transaction: it takes the payment, is responsible for collecting and remitting sales tax or VAT in the buyer's jurisdiction, issues the receipt, and handles chargebacks and returns. Software companies use one to avoid registering for tax in every country they sell into.

In WisprBee: Paddle.com is the Merchant of Record for all WisprBee orders. Your purchase is also subject to Paddle's buyer terms, and the subscription can be managed from the link in your Paddle receipt.

Want the practical version instead of the definitions? The how-to guides cover setting up your key, teaching it your names, and dictating into any app — and the FAQ answers the rest.

Put the keyboard down.

Hold a key, say what you mean, and let WisprBee write it right — for wherever it's going. Your fastest, cleanest writing is one keystroke away.

Download free See the 60-sec demo

Free to start · macOS · up to 4× faster than typing