Glossary: dictation, speech recognition and related terms explained

In shortDictation is when your computer turns spoken words into written text. It uses a speech recognizer (speech-to-text) that runs on your device or on a server; the result is a transcript, which a good tool cleans of filler words and punctuates. This glossary explains the key terms, grouped by topic.

What are the basic terms of dictation?

Dictation

Entering text by speaking: you say what you want to write, and the computer or phone writes it down. The word comes from the old office practice of dictating a letter to a typist; today, speech recognition software takes that role. You can dictate within a specific app (such as Word) or system-wide, into any text field.

Speech recognition (speech-to-text, ASR)

The technology that turns an audio recording into written text. It's also called speech-to-text or automatic speech recognition, ASR for short. Today's systems are neural networks trained on large amounts of recorded speech: rather than “looking up” each word in a dictionary, they estimate the most likely text from the whole stream of sound.

Voice typing

Another name for dictation. Windows (Windows + H) and Google Docs use this term; the Mac and Word call the same thing “Dictation”. In practice, these words all mean the same thing – only the menu names differ. More: Dictation on a computer – the complete guide.

Transcript

The raw output of speech recognition: text that records word for word what was said – filler words, repetitions and broken-off sentences included. A transcript isn't finished text yet; a good dictation tool turns it into a readable, punctuated version.

What happens behind the recognition?

Language model

A computer model that “knows” which words and sentences are likely in a language. In speech recognition, it helps decide between options that sound alike: “there”, “their” and “they're” sound the same, so the rest of the sentence decides which is right. That's also why a computer recognizes a sentence spoken fluently better than one spoken word by word or syllable by syllable – the context is missing.

Large language model (LLM)

A language model trained on huge amounts of text that doesn't just know likely words, but can also write, rephrase and summarize text. In dictation tools, it's typically used to tidy up the transcript afterwards: adding punctuation, leaving out filler words, splitting sentences.

Recognition accuracy, word error rate

The quality of speech recognizers is often measured with the word error rate (WER): how many words had to be replaced, inserted or deleted for the transcript to match what was actually said. The lower, the better. Accuracy varies by language, microphone and environment; a result measured in one language doesn't carry over directly to another.

Speaker recognition

Not to be confused with speech recognition. Speech recognition works out what was said; speaker recognition works out who is speaking. The latter is used for voice-based identification, or in transcripts of recordings with several people, where the transcript marks who said which sentence. A single-user dictation tool doesn't need it: all that matters is that your own words are written down accurately.

How does punctuation get into the text?

Punctuation command

A spoken word that the dictation tool doesn't write out but turns into a punctuation mark: “period”, “comma”, “question mark”, “exclamation mark”, “colon”. Line-break commands belong here too (“new line”, “new paragraph”). Exactly which words a tool understands varies by system, version and language – the vendor's help pages are the reference. Mac dictation, for example, inserts a mark when you say its name, and has separate commands for a new line and a new paragraph; Google Docs lists “exclamation point” where Apple lists “exclamation mark”.

Automatic punctuation

When the tool adds commas, periods and question marks without being told, based on the rhythm of your speech and the structure of the sentence. According to Apple, a Mac automatically inserts commas, periods and question marks in supported languages – English among them – and you can turn this off in Keyboard settings. Automatic punctuation is convenient but not infallible: a longer pause doesn't always mean the end of a sentence.

Special characters by voice

Beyond punctuation, some tools recognize other symbols too, such as “at sign” for @, which is handy when dictating email addresses. Check the finished text here as well, because the same words can appear in a sentence with their ordinary meaning.

What does text cleanup mean?

Filler word

Words and sounds that are common in speech but unnecessary in writing: “um”, “uh”, “so”, “like”, “you know”, “basically”. While speaking, they give you time to think, and listeners barely notice them; written down, they're distracting and make the text hard to read. A raw transcript keeps them; text cleanup leaves them out.

Text cleanup

Tidying up a transcript: removing filler words and unnecessary repetitions, adding punctuation and capital letters, splitting sentences – without changing what you meant. That's the key difference between cleanup and rewriting or paraphrasing: cleanup keeps your words and only makes them readable. Mondi's text cleanup, for example, does exactly this: it leaves out filler words, adds punctuation, and turns spoken “period”, “comma” or “at sign” into symbols.

Capitalization

Adding capital letters at the start of sentences and in proper names. Speech has no capital letters, so the software always has to work them out. It usually handles well-known names fine, but rare surnames and company names often need correcting.

Where is your voice processed?

Cloud processing

The audio is sent to a remote server, turned into text there, and the result comes back to your computer. The advantage is that servers can run larger, more accurate models than a laptop; the downside is that you need an internet connection, and your voice leaves the device. Windows voice typing, for example, needs an internet connection according to Microsoft, because it uses Azure's online speech services.

On-device processing

Recognition happens entirely on the computer or phone, and the audio never goes online. That's good for privacy and works offline; in return, the model's size is limited by the device. According to Apple, on Macs with Apple silicon, general dictation is processed on the device in many languages – whether that applies to you depends on your macOS version, your language and your Mac.

History

Many dictation tools keep the texts you've dictated so you can find or paste them again. An important difference is whether that history is stored on your computer or on a server, and whether the audio is kept along with the text. That's a separate question from where the transcript is made: a tool that uses cloud recognition can still keep your history locally.

What should you ask a tool?

When you choose a dictation app, read its privacy documentation: where is the transcript made, who does the processing, is the audio stored, and where does the history stay? With Mondi, for example, OpenAI transcribes the audio and Anthropic does the text cleanup; your audio isn't stored, and your history stays on your Mac.

What do you need to set up for the microphone?

Microphone permission

Modern operating systems grant microphone access app by app: a program can only hear you if you've explicitly allowed it. On a Mac, you can see and change these permissions for each app under Apple menu › System Settings › Privacy & Security › Microphone. If a dictation app “can't hear anything”, this permission is often what's missing – more: Mac dictation not working.

Input device

The microphone your computer is currently listening to: the built-in one, earbuds, a headset or an external USB microphone. If several are connected, the system may not use the one you're speaking into. The quality of dictation often depends on this setting rather than on the software.

Background noise and noise reduction

Background noise – conversation, music, a fan, traffic – reduces recognition accuracy, because the model has to separate your voice from its surroundings. Many microphones and systems use noise reduction, but it's no miracle cure: you get the best results in a quiet place, speaking close to the microphone.

Recording indicator

On a Mac, an indicator in Control Center shows when an app is using the microphone or has used it recently. That way you always know when your computer is listening.

What are voice commands and push-to-talk?

Voice command

A spoken instruction that the app carries out instead of writing it down: “Open Gmail”, “Remind me in twenty minutes”. The difference between a voice command and dictation is that one creates text and the other triggers an action. A good tool tells them apart and doesn't type the command into your text. Read about Mondi's voice commands here: Voice commands.

Push-to-talk

A mode where recording starts when you press a key or button and stops when you release it or press it again – like a walkie-talkie. The advantage is that you decide exactly when the computer listens, so nothing said in the background ends up in your text by accident. Mondi works this way: by default, you dictate into any app with the right ⌥ Option key.

Continuous dictation

The opposite of push-to-talk: once turned on, dictation keeps listening until you stop it or there's silence for a while. It's convenient for longer texts, but watch out for background speech slipping into your text.

What is accessibility?

Accessibility

Operating system features that help people with visual, hearing or motor impairments use a computer. On a Mac, they're under Accessibility in System Settings. Dictation and voice control belong here because they let people write and control the computer even if they can't or don't want to use a keyboard and mouse.

Voice Control (Mac), Voice access (Windows)

Controlling the whole computer by voice: opening menus, clicking buttons, scrolling, selecting and editing text. On a Mac, this is Voice Control, found under Apple menu › System Settings › Accessibility; on Windows, Voice access serves a similar purpose. They can do more than plain dictation, but learning the commands takes time, and their language support depends on the system version – check the vendor's current documentation.

What common errors might you run into?

Misrecognition

When the recognizer writes down a word that sounds similar but is wrong. In English, homophones (“write” and “right”), word boundaries (“a part” and “apart”) and short words that are easy to swallow (“an”, “and”, “in”) are typical. A misheard word is often a real word, so the spell checker won't flag it – only rereading will catch it.

Jargon, names, unusual vocabulary

Rare words – company names, product names, surnames, technical terms – are less familiar to recognizers, so they need correcting more often. Some tools help with a custom dictionary or word list.

Mixing languages

If you mix languages in a sentence – foreign names or terms in an English sentence, or English words in another language – the recognizer may spell them as if they belonged to the dictation language, or stumble. Check that dictation is set to the language you're mainly speaking, and review mixed sentences. If you're also curious about the speed difference: Dictation vs. typing.

Frequently asked questions

Back to Knowledge