You sit through a two-hour lecture, your notebook has three lines in it, and the exam is in three weeks. The recording exists — on your phone or in the university's lecture-capture system — but a recording is not study material until you can search it, summarise it and test yourself on it. That is what this pipeline does: record the lecture (with permission), turn the audio into a transcript, turn the transcript into grounded revision notes, and turn the notes into flashcards that come back on a schedule. Every step uses a free tool and keeps the AI tied to your own lecture material.
The short answer
- Ask first, record second. Most universities require the lecturer's permission, and in some countries recording spoken words without consent can be a criminal offence.
- Transcribe free. OpenAI's Whisper runs free on your own computer, or upload the audio to Gemini Notebook, which transcribes it on import.
- Summarise from the transcript, not from memory. Gemini Notebook answers only from the sources you upload, with citations.
- Check names, numbers and formulas against the slides. Transcripts mis-hear exactly the details exams test.
- Finish with flashcards in Anki. Generate cards from the verified notes, import them as a text file, and let spaced repetition schedule the reviews.
Step 0: ask before you press record
A recording captures the lecturer's voice and often classmates, which makes it personal data as well as the lecturer's intellectual property.
University rules come first. Policies differ, and your own university's policy is the one that decides. Glasgow's policy is a clear example: students must ask the course lecturer for consent in advance, say whether they want audio or video, and give as much notice as possible. Recordings are for personal study only, and recording after a lecturer has refused consent counts as a breach of the Code of Student Conduct. Some universities also restrict uploading recordings to third-party AI tools.
National law can bite too. In Germany, section 201 of the Criminal Code makes it an offence to record another person's privately spoken words without consent, punishable by up to three years in prison or a fine. Spain's data-protection rules, as explained by the Universitat Politècnica de Catalunya, require the speaker's express consent for recording a lecture and for communicating it afterwards. For the privacy side of putting course material into AI tools see article #42: GDPR and ChatGPT: What Students Should Never Paste Into AI (and How to Opt Out in 2 Minutes).
The practical rule is simple. Before the first lecture, ask the lecturer by email whether you may record for your own study and run the recording through a transcription tool. If the university already records the lecture, use the official recording: the permission question is already settled and the audio is better. Whatever permission you get, keep the recording to yourself: never share it, post it or sell it.
Step 1: record audio the AI can actually hear
A transcript is only as good as the audio, and lecture halls are hostile to microphones: echo, coughing, a lecturer who walks away from the podium. Sit near the front, near a speaker. That is where your phone picks up the clearest signal. Record in MP3 or WAV. MP3 at a normal bitrate keeps file sizes small without hurting accuracy. Test once. Record five minutes, transcribe it, and read the result — if it is mostly guesswork, change your seat. Keep one file per lecture, named clearly. "2026-11-20-Biology-101-Cell-Division.mp3" is findable in revision week; "recording-47.mp3" is not.
Step 2: turn the audio into a transcript
Two free routes:
| Option | Cost | How it works | The catch |
|---|---|---|---|
| Whisper on your own computer | Free, forever | OpenAI's open-source speech model, installed with one command and run locally | You install Python, PyTorch and ffmpeg yourself |
| Gemini Notebook audio upload | Free with a Google account | Upload the audio; it is transcribed at import and the text becomes a source | The audio goes to Google's servers, and usage is metered |
| Gemini Notebook mobile recorder | Free with a Google account | Record up to one hour in the app; the transcript becomes a source | One hour per recording; only the transcript is kept |
Whisper, the do-it-yourself route. Whisper is a speech-recognition model OpenAI released as open source under the MIT licence: the code and model weights are free to download and run from GitHub. Install it with `pip install -U openai-whisper`, plus ffmpeg, and transcribe a file with a single command. It comes in six model sizes, with a faster "turbo" version as the default. Whisper is a model, not an app, so there is no friendly interface — but the audio never leaves your machine. Its documentation warns that performance "varies widely depending on the language", so test it on one lecture first.
Gemini Notebook, the easy route. If you already use Gemini Notebook — the tool Google renamed from NotebookLM in July 2026 — transcription is built in. Upload the audio file as a source and it is transcribed at import; the transcript, not the raw audio, is saved as a source you can question and summarise. Google's help page says audio with no speech is not supported and low-quality imports may fail. The mobile app can record your voice for up to one hour and add the transcript straight to a notebook. If the recording contains other students' voices, run Whisper locally instead.
Step 3: from transcript to revision notes
A transcript is not study material. A two-hour lecture is roughly 15,000 spoken words. The job now is to compress it into notes you can use — without letting the AI add facts that were never said.
One notebook per subject. In Gemini Notebook, create a notebook for the course and add the lecture transcript plus the slides for the same lecture. Google's help page says each source can hold up to 500,000 words or 200MB, and a free account can include up to 50 sources per notebook, with up to 100 notebooks in total — far more than a degree needs.
Ask for a grounded summary. The technique is the same one this blog's lecture-notes guide uses see article #6: How to Use AI to Summarize Lecture Notes in 5 Minutes (Gemini Notebook): ask the notebook to summarise from the sources only, with citations. Good prompts are specific: "Summarise this lecture in 600 words, organised by the headings on the slides, and cite the source for each section." Then ask for the definitions in the exact wording the lecturer used and the examples given for each concept.
Use the Audio Overview as a second pass, not a first one. Gemini Notebook can turn your sources into a short podcast-style conversation between two hosts. Listen to it on the walk to your next lecture: hearing the material explained a second time, in different words, is genuine revision. Since September 2026, the free tier runs on a compute-based allowance that refreshes every five hours up to a weekly limit, and Audio Overviews are among the heaviest outputs, so generate the overview for the lecture you are about to revise, not for all twelve at once.
Step 4: check the transcript before you trust it
Transcription is pattern matching, not understanding, and it fails in exactly the places exams care about. Names, numbers and formulas are the weak points. A lecturer saying "Kierkegaard" becomes "care guard". "0.05" becomes "0.5" — a tenfold error in a significance level. A formula spoken aloud comes back as prose that no longer parses.
Models can also invent text that was never spoken. A 2024 study by Koenecke and colleagues (ACM FAccT 2024) ran more than 13,000 audio segments through Whisper and found that roughly 1% of transcriptions contained hallucinated phrases or sentences present nowhere in the audio — and that 38% of those hallucinations carried explicit harms, such as invented violence, made-up names or false medical details.
The verification pass takes ten minutes and is not optional. Open the transcript next to the slides and check four things: every name and term, every number and date, every formula, and every sentence that looks oddly fluent or oddly specific. If a claim in the summary has no citation, or the citation points to a passage that does not say that, delete it. A note you verified is worth more than a perfect-looking note you did not.
Step 5: turn verified notes into flashcards
Notes help you understand; flashcards help you remember.
Generate the deck from the verified notes, not from the raw transcript. In Gemini Notebook, open the Studio panel and choose Flashcards: you can set the difficulty, choose how many cards to make, and add a prompt saying what to focus on, such as "only the definitions and formulas from lecture 7". Generate one lecture at a time — smaller batches produce better questions than a whole semester dumped in at once. Then read every card: delete the vague ones, fix wrong answers, split double questions.
Move the keepers into Anki. Gemini Notebook has no spaced-repetition scheduler of its own: it can quiz you, but it will not bring a card back the day before you would forget it. That is Anki's job. Download the deck as a CSV file and import it into Anki as a text file — the workflow this blog's flashcard guide recommends see article #7: Best AI Flashcard Apps for Students: Memorize Faster with Spaced Repetition. Anki is free on Windows, macOS, Linux, Android and the web, with free sync through AnkiWeb; the only price in the whole product is AnkiMobile for iPhone and iPad, a one-time $24.99 on the US App Store. Its FSRS scheduler works out when each card is due, so your job is simply to clear the due queue every day.
Keep the loop running. After each lecture: record, transcribe, summarise, verify, make cards. Do it the same evening, while the lecture is still fresh enough for you to spot the transcript's mistakes. By revision week, every lecture already has notes and a deck — the work you do in week three is what saves you in week thirteen.
Frequently asked questions
Can I record a lecture without asking?
No — not safely. Your university's policy decides, most require the lecturer's permission in advance, and in countries like Germany recording someone's spoken words without consent can be a criminal offence. Ask by email before the first lecture.
Is Whisper really free?
Yes. Whisper's code and model weights are released under the MIT licence, so downloading and running them costs nothing. What you supply instead is setup: installing Python, PyTorch and ffmpeg, and the computer time to process the audio.
What if my lecture is longer than an hour?
Gemini Notebook's mobile recorder stops at one hour per recording, so split a long lecture into two recordings, or record the full audio on your phone and upload the file instead — uploaded files are limited to 200MB per source, which covers hours of MP3 audio.
How accurate are AI transcripts?
Good enough to work from, never good enough to trust blindly. Names, numbers and formulas are mis-heard most often, and a 2024 study found roughly 1% of Whisper transcriptions contained hallucinated sentences that were never spoken.
Can I upload a recording that includes other students asking questions?
Think twice. Classmates on a recording are identifiable people, and their words are their data too. If your permission covers only your own study, keep the file local and transcribe it with Whisper on your own computer.
Do I need to pay for any of this?
No. Whisper is free, Gemini Notebook's standard tier is free with a Google account, and Anki is free everywhere except the iPhone app. Pay only when a free limit has actually blocked you twice.
The bottom line
A lecture recording is raw material, not study material. The pipeline is simple: get permission, record clean audio, transcribe it free with Whisper or Gemini Notebook, summarise it from your own sources with citations, check every name, number and formula against the slides, and turn the verified notes into Anki flashcards. Do it the evening of each lecture. The AI does the typing; the checking and the remembering are still yours.
Related guides
- How to Use AI to Summarize Lecture Notes in 5 Minutes (Gemini Notebook) (article #6)
- Best AI Flashcard Apps for Students: Memorize Faster with Spaced Repetition (article #7)
- GDPR and ChatGPT: What Students Should Never Paste Into AI (and How to Opt Out in 2 Minutes) (article #42)
Facts here were checked against official pages in November 2026: Google's Gemini Notebook Help, OpenAI's Whisper repository on GitHub, Anki's official site, the University of Glasgow's Recording of Teaching Policy, the University of Leeds recording policy, and the Universitat Politècnica de Catalunya guidance. The Germany section 201 penalty and the Whisper hallucination figures come from secondary sources quoting the statute and the Koenecke et al. study. This is general information, not legal advice.