Echo · Glossary

Glossary of meeting and transcription terms

Plain definitions of the words used around meeting notes, speech-to-text and Arabic dialects.

Meetings

Action item

An action item is a specific task agreed in a meeting, assigned to a named person and usually given a deadline, so that a discussion turns into work.

An action item answers three questions: what needs to be done, who will do it and by when. For example, "Sara sends the revised budget to finance by Thursday" is an action item; "we should look at the budget" is not. Action items are the part of meeting notes people return to most, so they are usually listed separately at the end of the minutes or in a follow-up email and reviewed at the start of the next meeting.

Echo writes a to-do list from each recording alongside the summary.

Agenda

An agenda is the list of topics a meeting will cover, usually in order and often with a time and an owner for each item, shared before the meeting starts.

An agenda tells participants why the meeting exists and what they should prepare. A typical agenda has a goal, three to five items, the person leading each item and an estimate of time. For example: review last week's action items, decide on the launch date, any other business. Because minutes usually follow the agenda's structure, a clear agenda also makes notes easier to write and to read afterwards.

Decision log

A decision log is a running record of decisions a team or project has made, each with its date, the people involved, the reasoning and any alternatives considered.

Unlike minutes, which cover one meeting, a decision log collects decisions from many meetings and conversations in one place. A typical entry reads: "Use the existing vendor for phase one; decided by the project leads; cheaper and faster to start; revisit after launch." The log stops teams from reopening settled questions and helps new members understand why things are the way they are.

Follow-up email

A follow-up email is a message sent to participants after a meeting that confirms the main decisions, lists action items with owners and deadlines and shares any materials.

A follow-up email turns what people heard into something they can check. It is best sent soon after the meeting while details are fresh, and it is short: a line of thanks, the key decisions, the action items and the date of the next meeting if there is one. It also gives anyone who misunderstood a decision a chance to correct it in writing before work starts on the wrong assumption.

Meeting minutes

Meeting minutes are the official written record of a meeting: who attended, what was discussed, what was decided and who agreed to do what.

Meeting minutes are a structured record, not a word-for-word account. They usually list the date, attendees, agenda items, decisions and action items with owners and deadlines. A board or committee may approve the minutes at its next meeting, which gives them formal weight. Good minutes let someone who missed the meeting understand the outcome in a few minutes, and they settle later disagreements about what was agreed.

Echo's Standard template lays a note out as minutes, and you can export notes to PDF or Word.

See also: Action item Decision log Agenda · More

Meeting recap

A meeting recap is a short, informal summary of a meeting's outcome, covering the key points, decisions and next steps, written for people who need the result quickly.

A recap is lighter than formal minutes. It might be five bullet points in a team chat or the top of a follow-up email: what was agreed, what is still open and who is doing what next. Recaps suit internal and recurring meetings where nobody needs a formal record but everyone needs the same understanding. Many teams now generate a first draft of the recap automatically from a recording and then edit it.

Echo writes a summary with key points and decisions from each recording.

One-on-one meeting

A one-on-one meeting is a regular private conversation between two people, most often a manager and a team member, about progress, priorities, feedback and growth.

One-on-ones, often written 1:1, are usually short and recurring, for example thirty minutes every week or two. The team member often sets much of the agenda: blockers, career questions, workload or feedback in both directions. Because the conversation can be personal, notes are typically brief and shared only between the two people, focusing on agreed next steps rather than a full record. Ask before recording a one-on-one.

Stand-up meeting

A stand-up meeting is a short, usually daily team check-in where each person says what they did, what they will do next and what is blocking them.

Stand-ups come from agile software teams and are kept deliberately brief, often fifteen minutes or less; the name comes from the habit of standing so nobody gets comfortable. Each person answers the same three questions, and problems are noted for a separate conversation rather than solved on the spot. Notes from a stand-up are usually just a list of blockers and who will follow up on each one.

Speech and transcription

Automatic speech recognition (ASR)

Automatic speech recognition (ASR) is the field of computing, and the software it produces, that identifies the words in human speech and outputs them as text.

An ASR system takes audio, breaks it into short frames and uses statistical or neural models to predict the most likely sequence of words. Modern systems are trained on large amounts of transcribed speech, so they work best on the languages, accents and recording conditions they have seen. ASR is the first step in most voice products: a notetaker, for example, runs ASR to get a transcript, then other models summarise it.

Dictation

Dictation is speaking text aloud so that it is written down for you, today usually by speech-to-text software that types your words as you say them.

Dictation differs from transcription in intent: you are composing text on purpose, such as an email, a message or a note, rather than recording a conversation. Many dictation tools let you say punctuation, or add it automatically, and some tidy up filler words. It is useful when typing is slow, on a phone, while walking or for people who think faster than they type. Good dictation in Arabic needs to handle both dialect and Modern Standard Arabic.

Echo includes a keyboard for iPhone and Android that types what you say in any app.

Noise suppression

Noise suppression is audio processing that reduces background sounds such as traffic, fans, typing or crowd chatter while keeping the speaker's voice clear.

Noise suppression can use classic signal processing or neural networks trained to separate speech from noise. Video-calling apps and many headsets apply it automatically. It helps both listeners and speech recognition, but it is not magic: heavy suppression can distort voices or cut quiet words, and it cannot recover speech that was never captured clearly. Placing the microphone close to the speaker in a quieter room usually improves a transcript more than any filter.

Speaker diarization

Speaker diarization is the process of splitting an audio recording by speaker, so that each segment of a transcript is labelled with who was talking.

Diarization answers "who spoke when?", not "what was said". The system groups segments by voice characteristics and labels them Speaker 1, Speaker 2 and so on; a person or a calendar invite can then attach real names. It is harder when people interrupt each other, speak over one another, sound alike or share one microphone. Accurate speaker labels matter for minutes, because decisions and action items depend on knowing who said them.

Speech-to-text

Speech-to-text is technology that converts spoken words in audio into written text, either live as someone speaks or afterwards from a recording.

Speech-to-text, often shortened to STT, is the everyday name for what engineers call automatic speech recognition. It powers voice typing on phones, live captions, voice assistants and transcripts of meetings and lectures. Results depend on audio quality, background noise, accents, how many people talk at once and whether the system was trained on the language and dialect being spoken, which is why Arabic dialects remain harder than standard English for many systems.

Timestamp

A timestamp in a transcript is a time marker, such as 00:14:32, that shows where in the recording a word, sentence or speaker turn occurs.

Timestamps link the written text back to the audio. They may appear at every speaker change, at fixed intervals or for each word. In practice they let you jump to the exact moment a decision was made, check a quote against the recording or cite a point in an interview. When a transcript contains a doubtful word, the timestamp is the fastest way to listen again and correct it.

Echo's transcripts include timestamps.

Transcription

Transcription is the process of turning recorded or live speech into written text, done by a person, by software, or by software followed by human review.

The result is called a transcript. Transcription can be verbatim, keeping every filler word and false start, or clean, lightly edited for readability. Transcripts often add speaker labels and timestamps so readers can find who said what and when. For meetings, the transcript is the raw material: summaries, minutes and action items are all drawn from it, so errors in the transcript carry into everything built on top.

Echo produces a full transcript with timestamps for each recording.

Verbatim transcript

A verbatim transcript is a word-for-word written record of speech that keeps exactly what was said, including repetitions, filler words, false starts and often non-speech sounds.

In a verbatim transcript, "So, um, we, we need the report by, uh, Monday" stays as spoken; a clean transcript would read "We need the report by Monday." Verbatim transcripts matter where exact wording counts, such as interviews, research and some legal or compliance settings. For everyday meeting notes, a clean transcript or a summary is usually easier to use, while the verbatim version stays available for checking a quote.

Voice activity detection (VAD)

Voice activity detection (VAD) is a technique that decides, moment by moment, whether an audio signal contains human speech or only silence and background noise.

VAD is a small but essential step before speech recognition. By marking where speech starts and stops, it lets a system skip long silences, split audio into manageable segments and avoid "transcribing" an air conditioner or keyboard clicks as words. Phones and calling apps also use it to save bandwidth. When VAD is too aggressive it can clip the first syllable of a sentence or miss quiet speakers, which shows up as missing words in a transcript.

Word error rate (WER)

Word error rate (WER) is the standard measure of speech recognition accuracy: the share of words a system got wrong compared with a correct reference transcript.

WER adds up substitutions (wrong words), deletions (missed words) and insertions (extra words), then divides by the number of words in the reference. Lower is better. If a ten-word sentence comes back with one wrong word and one missing word, the WER is 20 percent. WER is useful for comparing systems on the same audio, but it treats every word equally and is sensitive to spelling choices, which matters for Arabic dialects that have no single standard spelling.

Arabic

Arabic dialect

An Arabic dialect is a spoken regional variety of Arabic, such as Gulf, Levantine, Egyptian or Maghrebi, used in everyday life rather than formal writing.

Arabic dialects, اللهجات or العامية, differ from Modern Standard Arabic and from each other in pronunciation, vocabulary and grammar. "What do you want?" is شو بدك؟ in Levantine and عايز إيه؟ in Egyptian, for example. Dialects have no single standard spelling, and speakers often mix them with MSA and English. That is why dialect support is the main test of an Arabic transcription tool.

For Arabic, Echo covers Gulf, Levantine, Egyptian, Maghrebi and Modern Standard Arabic.

Arabizi

Arabizi is a way of writing Arabic, usually dialect, in Latin letters and numerals, common in text messages and social media; it is also called Franco-Arabic.

Arabizi uses digits for Arabic sounds that have no Latin letter: 3 for ع, 7 for ح, 2 for ء and 5 for خ, so حبيبي becomes "7abibi". Spelling is informal and varies between people and countries. Arabizi matters for notes because some people write their own meeting notes or search their transcripts this way, while most transcription tools output Arabic script.

Code-switching

Code-switching is the practice of moving between two or more languages or language varieties within one conversation, often within a single sentence.

In Gulf offices, a sentence like خلّنا نسوي follow-up بعد الـmeeting ("let's do a follow-up after the meeting") is ordinary speech. Speakers also switch between dialect and Modern Standard Arabic, or between Arabic and French in North Africa. Code-switching is hard for transcription because the system must recognise both languages and write each in its own script, instead of forcing English words into Arabic letters or dropping them.

Echo handles recordings that mix Arabic with English.

Egyptian Arabic

Egyptian Arabic, or Masri, is the Arabic dialect spoken in Egypt and one of the most widely understood dialects in the Arab world.

Egyptian Arabic, المصرية, is known across the region through Egyptian film, television and music. Common examples include إزيك؟ ("how are you?"), عايز ("I want") and دلوقتي ("now"). In Cairo the letter ج is pronounced as a hard "g" and ق usually as a glottal stop. These sound changes mean a system trained only on Modern Standard Arabic may mishear common Egyptian words.

Echo covers Egyptian Arabic.

Gulf Arabic

Gulf Arabic, or Khaleeji, is the group of Arabic dialects spoken in the countries of the Arabian Gulf, including the UAE, Saudi Arabia, Kuwait, Qatar, Bahrain and Oman.

Gulf Arabic, الخليجية, varies from country to country and even city to city, but shares features such as شلونك؟ ("how are you?") and وايد ("a lot"), and in many areas the letter ك is pronounced like "ch". In Gulf workplaces, meetings often mix Gulf Arabic with English and other Arabic dialects, so notes and transcripts need to handle all three.

Echo covers Gulf Arabic, including recordings that mix Arabic with English.

Levantine Arabic

Levantine Arabic, or Shami, is the group of Arabic dialects spoken in Syria, Lebanon, Jordan and Palestine, widely understood across the Arab world.

Levantine Arabic, الشامية, is familiar to many Arabic speakers through television drama and music. Common features include شو ("what"), هلّق ("now") and كتير منيح ("very good"), and in cities the letter ق is often pronounced as a glottal stop. Levantine speakers in business settings frequently mix in English or French words, which a transcript needs to keep rather than force into Arabic.

Echo covers Levantine Arabic.

Maghrebi Arabic

Maghrebi Arabic is the group of Arabic dialects spoken in North Africa, mainly Morocco, Algeria, Tunisia and Libya, often called Darija.

Maghrebi Arabic, الدارجة المغاربية, is usually the hardest family for speakers from the Gulf or the Levant to follow, because of its short vowels, Amazigh influence and many French loanwords. Moroccan examples include بزاف ("a lot") and دابا ("now"). In offices, speakers often switch between Darija and French within one sentence, so a transcription tool needs to handle both languages, not Arabic alone.

Echo covers Maghrebi Arabic, including recordings that mix Arabic with English.

Modern Standard Arabic (MSA)

Modern Standard Arabic (MSA) is the shared formal variety of Arabic used across the Arab world in writing, news, education, official documents and formal speeches.

MSA, الفصحى in Arabic, descends from Classical Arabic with modernised vocabulary. It is understood across countries, but it is rarely anyone's everyday spoken language; people grow up speaking a local dialect and learn MSA at school. In a business meeting, a speaker may open a presentation in MSA and switch to dialect for discussion. Speech systems trained mainly on MSA, such as news audio, often struggle with that everyday dialect speech.

Echo covers Modern Standard Arabic as well as Gulf, Levantine, Egyptian and Maghrebi Arabic.

AI notetakers

AI notetaker

An AI notetaker is software that records a meeting, lecture or call, transcribes it with speech recognition and uses AI to produce summaries, action items and searchable notes.

AI notetakers replace or support manual note-taking so participants can focus on the conversation. They capture audio either by sending a bot into a video call or by recording on the user's own device, then generate a transcript and, from it, a summary, decisions and to-dos. Many also let you ask questions about past meetings. Important differences between them include language and dialect support, how audio is captured, privacy and price.

Echo is an AI notetaker for iPhone, Android and Mac, made by CNTXT, that transcribes Arabic and English: every Arabic dialect, and both mixed in one recording.

AI summary

An AI summary is a condensed version of a transcript or document written by a language model, highlighting the main points, decisions and next steps.

For meetings, an AI summary typically turns an hour of conversation into a short overview, a list of key points, decisions and action items. Its quality depends on the transcript underneath: misheard names or dialect words can lead to wrong conclusions. Language models can also state things that were never said, so a summary should be checked against the transcript before it is shared or used for decisions, especially for numbers, names and commitments.

Echo writes a summary with key points and decisions, plus a to-do list, from each recording.

Ask your notes (retrieval over notes)

Ask your notes is a feature that answers natural-language questions using your own transcripts and notes, by retrieving the relevant passages and having an AI model answer from them.

Instead of rereading transcripts, you ask something like "What did we agree about the delivery date?" The system searches your notes for relevant passages, a technique known as retrieval-augmented generation, and a language model writes an answer based on them, ideally pointing to the source. Answers are only as good as the notes and the retrieval: if a topic was never recorded or was transcribed wrongly, the answer may be incomplete, so checking the cited passage is good practice.

Echo's "Ask Echo" answers questions from one note or across all your notes.

Meeting bot

A meeting bot is an automated participant that joins a video call, such as Zoom, Google Meet or Microsoft Teams, to record, transcribe or take notes.

A meeting bot appears in the participant list, usually with a name like "Notetaker", and captures the call's audio and sometimes video from the meeting platform. It can join calls the user is not attending and is visible to everyone, which some people see as transparency and others find intrusive. Some organisations block unknown bots, and hosts may need to admit them. The alternative is recording on the user's own device without adding a participant.

Echo for Mac records from the menu bar using the microphone and system audio, so no bot or extra participant joins the call.

On-device recording

On-device recording means capturing a meeting's audio on the user's own phone or computer, instead of through a bot or service that joins the call.

With on-device recording, the app listens through the device's microphone and, on a computer, can also capture the audio coming out of it. It works the same for in-person meetings and for calls on any platform, and nothing new appears in the participant list. Because no one sees a bot, the person recording is responsible for telling others and asking before recording. On-device recording describes how audio is captured, not necessarily where it is processed.

Echo records on iPhone, Android and Mac; on Mac it uses the microphone and system audio from the menu bar.

System audio

System audio is the sound a computer itself plays, such as other participants' voices in a video call, as opposed to sound picked up by the microphone.

In an online meeting, the microphone captures your voice and the system audio carries everyone else's. A recording that takes only the microphone may miss remote participants, especially if you wear headphones; capturing system audio as well gives both sides of the call. Operating systems usually require the user to grant permission before an app can capture system audio, and the same consent expectations apply as for any recording.

Echo for Mac records the microphone and system audio from the menu bar, and works with Zoom, Google Meet and Microsoft Teams.

Echo writes these notes for you from a recording. Free for 30 minutes a month.

Try Echo free