How to transcribe a lecture into notes

A recording of a lecture is only useful once you can read it, search it and study from it. Here is a workflow from asking permission to reviewing before the exam.

Last checked:

Writing notes during a lecture forces you to choose between listening and writing. Recording the lecture and transcribing it afterwards lets you do the listening in the room and the writing later, at your own pace. This guide covers the whole path: permission, a recording that is good enough to transcribe, the transcription itself, and turning pages of raw text into notes you can learn from.

1. Ask before you record

Many universities have rules about recording lectures, and many lecturers have their own. Some record and publish their lectures already; some are fine with students recording for personal study; some say no, for example when guests speak or students discuss personal matters.

  • Ask the lecturer, ideally before the first lecture, and mention that the recording is for your own study.
  • Check your university’s policy. Search its website for “lecture recording” or “recording policy”. Students who need recordings as a disability accommodation usually have a separate route through the disability or accessibility office.
  • Keep it to yourself. A recording made for your own study is not something to share or post.
  • Stop when asked, for example during a discussion where classmates share personal views.

Ask before the first session. Recording without asking is a poor way to find out whether they would have said yes.

2. Get a recording that transcribes well

Transcription software is only as good as the audio you give it. Echo, distance and noise cost more accuracy than the choice of tool.

In a lecture hall:

  • Sit near the front, close to the lecturer or to a loudspeaker if the lecture is amplified. Distance is the biggest enemy of a clean recording.
  • Keep the laptop or phone still. Typing, rustling papers and bumping the desk are loud next to the microphone. A phone on a folded jumper picks up less desk noise than one lying flat on the table.
  • Point the microphone at the lecturer. On a laptop, that usually means the screen facing the front of the room.
  • Try a short test in the first minutes: record 30 seconds, listen back with headphones, and move if you cannot understand the words clearly.
  • Check battery and storage before a two-hour lecture. Audio files are small, but a laptop that goes to sleep stops recording.

Online lectures: record the computer’s audio, not your microphone picking up your speakers. On a Mac that needs a little extra setup, which our guide on recording system audio on a Mac explains. If the course platform already publishes recordings or captions, use those; they are usually cleaner than anything you record yourself.

3. Choose a way to transcribe

There are three broad options, and they differ mainly in accuracy, cost and where your audio goes.

Built-in dictation and voice memos

macOS Dictation turns speech into text as you speak, but it is made for dictating, not for transcribing a recording. Apple notes that it stops automatically when no speech is detected for 30 seconds, and that it is not available in all languages or regions (Apple’s Dictation guide).

The Voice Memos app can show a transcript of a recording on a Mac with Apple silicon running macOS Sequoia or later, but it is not available in every country, region or language (Apple’s Voice Memos guide). If it covers your language, it is the quickest way to try this at no cost.

Whisper-based tools on your own computer

Whisper is a speech recognition model from OpenAI whose model weights are public, which means many apps can run it on your own computer. You can run it for free from Terminal with a tool such as whisper.cpp, or use an app that wraps it in a normal window. Your audio never leaves the machine, and there is no monthly limit on hours.

The trade-off is speed and setup. Larger models are more accurate and slower; on an Apple silicon Mac even a large model runs faster than real time. Our guide to running Whisper on a Mac walks through it.

Cloud transcription services

Many web services and apps upload your audio and send back a transcript, often with speaker labels and a summary. They are easy to use and usually accurate in widely spoken languages. Before you choose one, check:

  • Free tier limits. Free plans typically cap minutes per month or per recording, which a semester of two-hour lectures can use up quickly.
  • Where the audio goes and whether it may be used to train models. Your university may have rules about uploading lecture recordings to third parties.
  • Your language. Accuracy varies a lot between languages; test with one lecture before you commit.

Whichever you pick, set the language explicitly rather than relying on automatic detection, especially if the lecture is not in English.

4. Turn the transcript into study notes

A transcript is not notes. A two-hour lecture can run to well over ten thousand words of rambling, repeated spoken text. The work is in cutting it down.

Structure it first. Skim the transcript and mark where the topic changes: “Now let us turn to…”, “The second reason is…”, “Okay, a quick break”. Those become your headings. Many lecturers follow their slides, so the slide titles are a good skeleton.

Pull out the key terms. For each section, list the terms the lecturer defined and write each definition in one sentence, in your own words. Keep the exact quote only when wording matters, such as a legal definition or a formula.

Write the questions. For every section, write two or three questions that the section answers: “What are the three conditions for…?”, “Why does X fail when…?”. These become your revision questions later, and writing them forces you to understand the material rather than copy it.

Mark what the lecturer flagged. Search the transcript for phrases like “this will be on the exam”, “important”, “remember” and “a common mistake”. Lecturers say these out loud far more often than they put them on slides.

Keep a pointer back to the recording. Next to each heading or tricky point, note the time in the recording (for example 42:10). When a note does not make sense in three weeks, you can listen to that minute instead of the whole lecture.

A simple layout that works in any notes app:

# Lecture 5: Market failure (14 Oct)

## Externalities (05:30)
- Externality: a cost or benefit that falls on someone outside the transaction.
- Example: factory pollution affecting nearby residents.
- Q: Why does a market over-produce goods with negative externalities?

## Public goods (38:15)
- Non-rival and non-excludable.
- "This distinction comes up every year" (41:02)
- Q: Is a lighthouse a public good? What changes if it charges ships?

## To check
- Missed the second example at 52:40. Relisten.

If you use an AI tool to summarise the transcript, treat the summary as a first draft. Check it against the transcript, because summaries can drop conditions, merge two ideas into one or state something the lecturer did not say.

5. Review, do not just file

Notes that are never opened again do not help much. A few habits make the transcript worth the effort:

  • Tidy the notes within a day or two, while you still remember the lecture.
  • Answer your own questions without looking, then check. The ones you get wrong are what to study.
  • Search across lectures when revising: a keyword search across all your transcripts finds every time a concept came up, which is often faster than rereading.
  • Relisten selectively, using the times in your notes, instead of replaying whole lectures.

Where Hearkeeper fits

Hearkeeper is a Mac app that is not released yet. It records lectures in the hall with your Mac’s microphone and online lectures from your computer’s audio, with no bot joining the call. Every recording starts with a consent step, and the recording pill stays on screen so it is always visible that you are recording.

It transcribes on the Mac with whisper.cpp, with live captions during the lecture and a more accurate final pass afterwards, in 99 languages. A local language model then writes a summary and lets you ask questions across all your lectures, with every answer linking to the second it was said. Nothing is uploaded, and notes export to Markdown. The lecture recorder page has more.

If that sounds useful for next term, join the waitlist and we will email you when there is a build for you.

Try Hearkeeper first

Private meeting and lecture notes, made on your own Mac. Join the waitlist and we’ll email you when it’s ready.

One email when it’s ready, nothing else. Leave any time. Privacy