How to transcribe a lecture recording with speaker labels
Get an accurate, speaker-labelled transcript of any lecture recording — audio or video, in dozens of languages.
The short answer
To transcribe a lecture recording with speaker labels, upload the audio or video file and let a diarizing speech-to-text model process it: diarization is the step that tags each turn as a distinct speaker, so a seminar reads as a script rather than a wall of text. AuroraNote accepts m4a, mp3, wav and flac audio as well as mp4, mov and webm video. For video, the audio track is extracted inside your browser first and normalised to a 64 kbps mono MP3 — about 28 MB per hour — so only that small file is uploaded and the video never leaves your device. The language is detected automatically, including lectures where the speaker switches between languages mid-sentence, so you do not choose one up front. A 60 to 120 minute recording is processed in the background in a few minutes. Your first document is free with no card; after that transcription costs one credit per minute, charged only when it succeeds.
A transcript is the foundation of every other study step: you cannot summarise, map or quiz a recording until it is text. The good news is that modern speech-to-text is fast and accurate enough that transcribing a full lecture takes a couple of minutes, not the length of the recording.
Here is how to get a clean, speaker-labelled transcript from any lecture — whether you recorded audio on your phone or captured a video call.
How to transcribe a recording
- 1
Upload the recording
Add an audio file (m4a, mp3) or a video file. For video, the audio is extracted in your browser first so the video stays on your device.
- 2
Let AI transcribe with diarization
The recording is transcribed with automatic language detection and speaker labels — "Speaker 1", "Speaker 2" — which is ideal for seminars and Q&A.
- 3
Read, clean or export
Read the transcript in the workspace, switch to Clean Text to remove filler, or export to DOCX/Markdown.
What is speaker diarization, and why does it matter for lectures?
Speaker diarization is the step that answers "who spoke when": the model segments the audio by voice and tags each turn as Speaker 1, Speaker 2 and so on. It is a separate problem from transcription — a system can get every word right and still produce an unusable wall of text if it does not know where one person stops and the next begins.
For lectures the difference is practical. In a seminar you can tell the professor’s answer from a student’s question; in a group project you can attribute a decision to the right person. It matters most for exam prep, because the professor’s asides ("this will be on the exam") and clarifications are often the highest-value seconds in two hours, and labels make them findable.
What files can you transcribe, and what happens to them?
Audio files are uploaded as they are. Video files are handled differently, and this is the part worth understanding before you upload a lecture recorded in a room full of people: the audio is separated from the video inside your browser, using a WebAssembly build of ffmpeg, and normalised to a 64 kbps mono MP3 at 22.05 kHz. Only that file is sent. The video itself is never uploaded, which is also why a 2 GB recording does not need a 2 GB upload.
| You upload | Handled by | What leaves your device |
|---|---|---|
| m4a, mp3, wav, flac | Normalised in-browser, then uploaded | A mono MP3, around 28 MB per hour |
| mp4, mov, webm | Audio track extracted in-browser | Audio only — the video stays local |
| A 2-hour lecture | Processed in the background | Roughly 56 MB, not several GB |
| A PDF or slide deck | OCR, not transcription | The document itself |
Can it handle a two-hour lecture in more than one language?
Yes to both. Full lectures run 60 to 120 minutes and are processed in the background without you splitting the file manually, so you can keep working while it runs. Language detection is automatic, so you do not pick one up front, and a lecture where the professor switches language mid-sentence or quotes a source in another is handled in a single pass rather than producing a block of nonsense at the switch.
Output language is a separate choice from input language. The transcript follows the recording, but notes, summaries and mind maps generated from it can be produced in any of the 46 languages in the picker — which is the case for a lecture heard in one language and revised in another.
Frequently asked questions
How long can the recording be?
Full-length lectures of one to two hours are supported. Very long recordings are processed in the background so you can keep working.
Is transcription free?
Your first document is free with no card required, and a new account starts with 50 welcome credits. Transcription costs one credit per minute, so those 50 credits cover roughly a 50-minute lecture. Credits are charged only when a transcription succeeds — a failed or timed-out job costs nothing.
How accurate is it?
Accuracy depends mostly on audio clarity. With a reasonable recording, transcription is highly accurate and includes punctuation and paragraphing automatically.