How do you transcribe a lecture recording?

To transcribe a lecture, record it, run the recording through a speech recognition tool, then read the draft while you listen to the audio. A free tool on your own computer handles many lectures. A cloud service suits a very long recording, and a human transcriptionist is worth paying for only when a mistake would cost more than the fee. Lectures have three features that change the job: they run long, they use specialist terms, and the speaker is often some distance from the microphone. Plan for each one before you press record.

Accuracy on lectures depends on the tool, so test it on your own recording

A 2024 study ran eleven speech recognition services on university lecture recordings. Whisper large-v2, an open model from OpenAI, had an average word error rate of 2.9 percent on the English recordings ¹. Word error rate is the share of words a tool gets wrong, as explained in the glossary entry on word error rate. Google's average on English was 20.1 percent ², so the same lectures produced far more errors from one service than from another.

The authors found that accuracy varied widely between vendors and between individual recordings ³. The study also ran in 2024, and cloud services change, so treat these figures as a reason to test rather than a ranking. Run five minutes of your own lecture through the tool you plan to use, then check how many words are wrong before you process the whole course.

Free tools are enough for many lectures, but check the course rules first

Running Whisper on your own computer is one free route, and the audio stays on your machine. Built-in options, such as the free transcription in Apple's Voice Memos on a Mac, are covered with their requirements in our guide to how to transcribe audio to text, part of our transcription overview.

Check the rules before you record anything. Some institutions restrict these tools. The University of Birmingham's 2019 to 2020 guidance says teaching staff may prohibit transcription tools in some learning activities when they consider it necessary to protect academic integrity ⁴. That guidance is one university's rules for one year, so ask your own department in writing and keep the answer.

Long lectures need a file-size check and a cut between sentences

The OpenAI transcription API accepts files up to 25 MB ⁵, and a long lecture can exceed that. Compress the file first. If you still need to split it, cut between sentences. OpenAI advises against splitting in the middle of a sentence, because that can remove context and reduce accuracy ⁶. After you join the transcripts, check the first few words after each cut against the audio.

Names and terms go in a short prompt, and the quiet stretches get checked

Lectures are full of course names, formulas and the lecturer's own terms. Whisper's initial prompt can give the model custom vocabulary or proper nouns, which makes those words more likely to be transcribed correctly ⁷. Keep the prompt short and accurate: the course name, the lecturer's name and a dozen key terms. It is a hint, not a dictionary, so a long list can pull the output toward those words. See custom vocabulary for how this works in other tools.

Lectures also have long pauses, slide changes and questions from the room. Whisper can invent text when there is no speech. In a 2024 study of a 2023 version of Whisper, on recordings from people with and without aphasia, roughly 1 percent of transcriptions contained entire hallucinated phrases or sentences that did not exist in the audio ⁸. Read the quiet stretches first, and delete any phrase that does not match the sound.

What to do: five steps from recording to revision notes

1. Record close to the speaker. A microphone nearer the lecturer captures more of the voice and less of the room. Save the file in a format your tool accepts.
2. Check the rules. Confirm in writing that transcription tools are allowed for this course.
3. Prepare the file. Compress it if it is over the upload limit, and split it only between sentences.
4. Transcribe with a short prompt. List the course name, the lecturer and the key terms, and set the language yourself if you know it.
5. Proofread against the audio. Correct names, formulas and dates first, then the quiet stretches. Save the corrected transcript, then write your own summary of the main points to study from.

The last step matters for learning. A transcript records what was said, and it is useful for search and review. It does not replace notes you write yourself. For a lecture in a noisy hall, see how to transcribe low-quality audio. For a seminar with several speakers, see how to transcribe audio with multiple speakers. If the recording must stay on your device, read how to transcribe audio offline.

Yap, which we make, is an on-device dictation app for Mac and Windows. It types what you say into the app you are using, so it suits dictating your own notes. It does not transcribe a finished lecture recording.

For live captions in the room, the accuracy is a separate question. Can Whisper transcribe in real time explains why a live setup works differently from a batch transcript of a finished file.