Share the meeting tab
Click Record, pick the tab where the call runs and keep “Share tab audio” on. Your microphone is recorded separately, so the transcript knows who spoke: Me or Others.
Meeting Transcriber
Turn a meeting in a Chrome or Edge tab (Google Meet, or Teams and Zoom on the web) into a transcript with two speakers: Me and Others. Speech recognition runs on your laptop, and nothing is uploaded.
Fast gives quicker, rougher text. Accurate is better, especially in Dutch, Russian and Finnish, but slower: on an older laptop its text can arrive after the meeting.
Model not downloaded yet. Download it before the meeting.
Chrome will ask which tab to share. Pick the meeting tab and keep “Share tab audio” on. Then allow the microphone, so your own voice is included.
Record on your phone with any voice recorder app, then transcribe the file here. File transcripts have no Me or Others labels.
The transcript appears here while you talk.
Runs the Accurate model over this meeting's saved audio. It can take a while; keep this tab open.
Saved in this browser
No saved meetings yet.
Stored in this browser only. Clearing this site's data deletes it.
How it works
Speech becomes text on your laptop
The page cuts the audio at pauses and runs Whisper, an open-source speech recognition model from OpenAI, inside your browser. The model downloads once and stays cached.
Copy, download or come back later
Copy the text or download it as .txt or .md. Meetings stay in this browser until you delete them, and you can re-transcribe one with the Accurate model afterwards.
Limits
- Browser meetings only. The Zoom and Teams desktop apps cannot be captured; join from the web version instead.
- Chrome or Edge on a computer. Safari, Firefox and phones cannot share a tab's audio.
- The first use downloads the speech model: about 210 MB for Fast and about 590 MB for Accurate. After that it is cached.
- Use headphones. With speakers, the other side's voice can leak into your microphone and show up as Me.
- Two speakers only: Me is your microphone, Others is everyone in the tab. It does not tell the other participants apart.
- Live text is a draft. Its quality depends on your laptop, and the Fast model makes more mistakes in Dutch, Russian and Finnish than in English.
- Audio files: up to 2 hours each, and their transcripts have no Me or Others labels.
What leaves your laptop
- Audio and text: nothing. Both are processed and stored in this browser.
- The speech model: downloaded once from Hugging Face, the public model host.
- Analytics: the site's usual analytics, and only if you accept cookies. The tool area is masked, so session recordings never capture its text.