Private by design
Private AI meeting transcription with speaker labels
Free, and it turns recorded voice to text entirely in your browser: your audio never leaves your device. No signup, no subscription, nothing uploaded. Built for confidential work.
The first time you use it, it takes a minute or two to set up. After that setup is quick. Works with MP3, WAV, M4A, OGG, FLAC and AMR. Long recordings are handled in parts, so most files are fine whatever their length.
Refine with your AI is the best way: it gives you a prompt for your AI of choice, such as ChatGPT, Claude or Gemini, which corrects words using the context of the whole conversation and catches far more, like “our repark” for RV park. You approve every change either way.
How is this private?
Most transcription services upload your recording to their servers to process it. MeetingScribe does not. When you press Transcribe, the speech-recognition model is downloaded into your browser once, and then your audio is processed right here on your own device.
Your audio file, the transcript, and the meeting-notes prompt all stay on your computer. There is no account, no upload, and no copy of your recording sitting on someone else's server. There is no paid tier and no server that processes your content: the whole tool is the page in front of you. You can verify it: open your browser developer tools, watch the Network tab, and you will see the model download but never your audio going out.
If you paste the notes prompt or the Refine prompt into an outside AI chat, that is your own action on a service you choose, and only the text you paste goes there. This is designed for confidential workflows where uploading a recording is not an option. We do not claim any specific regulatory certification; you remain responsible for your own compliance obligations.
On your device
Transcription and speaker separation run locally in your browser.
No upload
Your audio is never sent to us or anyone else.
Yours to keep
Download the text, subtitles, and notes prompt. Nothing is stored by us.
Made for confidential work
If you cannot or will not upload a recording to a cloud service, this is for you.
Lawyers
Client interviews and privileged calls that cannot leave your control.
Learn moreTherapists
Session recordings that must stay confidential.
Learn moreJournalists
Source interviews you need to protect.
Learn moreEveryone else
HR, doctors, researchers, and anyone handling sensitive audio.
Learn moreNew here? See what you can and cannot do.
What you get
One transcription produces everything below, and all of it stays on your machine.
Mis-heard names and jargon, corrected with your own AI
Speech models hear sounds, not meaning, so the words that matter most, names, places and trade terms, are the ones they get wrong: on a real call, “hand dig” came out as “hand date” every time. Refine with your AI gives you a ready-made prompt for ChatGPT, Claude or Gemini, which corrects words using the context of the whole conversation. Paste its answer back and the page lists every changed word to keep or reject, with undo. Quick refine does a lighter pass with the AI built into Chrome, without anything leaving your computer. The guide to fixing mis-heard words shows what each one catches.
A transcript you can actually edit
The text appears in an editor in the page, so you can fix misheard names and technical terms before you download anything. Speech models mishear proper nouns more than ordinary words, and correcting them once here beats re-running the whole file.
Transcription with speaker labels
A second model separates voices and labels each turn, and you can rename the labels to the people who were in the room. It works best with one to three voices, and it does not identify anyone: it only tells voices apart inside the one recording. More on how speaker labels work.
Phone call recordings, including AMR
Call-recorder apps on Android often save AMR, including wideband AMR-WB from HD Voice calls, which no browser can play. MeetingScribe brings its own AMR decoders and runs them in the page, and audio that measures as phone-quality gets a lift to the consonant frequencies before transcription, which on a real 21-minute call recovered industry terms the untreated audio got wrong. Transcribing a call recording.
Subtitles with timestamps
Every run also exports SRT and VTT files, timed to your audio, for a video editor, YouTube, a desktop player, or a web page. See the guide to SRT and VTT for which format each tool expects.
Meeting notes, on your terms
One click copies a ready-made prompt containing your transcript and clear instructions, so any AI chat can turn it into a summary with action items. We deliberately did not build a one-click cloud summary, because it would send your transcript to a server and quietly undo the point of the tool. That reasoning is on the about page.
Common questions
- Is it really free, and what is the catch?
- Yes. This is free transcription without a subscription: no account, no file limit, no trial that expires, and no paid tier at all. The site is funded by advertising, which only loads if you accept cookies. The honest catch is what it can do, not what it costs: the first run takes a minute or two to set up, very long recordings in compressed formats can be more than a browser tab will hold (saving as WAV avoids that), and on really difficult audio a big paid service will still beat it.
- How can transcription work without uploading my file?
- The speech-recognition model is downloaded into your browser once and then runs on your own processor or graphics card. The model travels to your audio instead of your audio travelling to the model. After that first download it is cached, and transcription works even with your internet disconnected.
- What file formats can I use?
- MP3, WAV, M4A, OGG, FLAC, and AMR. AMR files from phone recorders, narrowband or wideband (AMR-WB, from HD Voice calls), which no browser can decode natively, work here because the tool loads its own AMR decoders and runs them in your browser. For video, export the audio track and transcribe that.
- Can it tell me who said what?
- Yes. Speaker separation runs on your device alongside the transcription and labels each turn, and you can rename the labels to real names. It works best with one to three voices, and it identifies nobody: it only tells voices apart within the one recording.
- Can I get subtitles?
- Yes. Every transcription also produces SRT and VTT subtitle files with timestamps, ready for a video editor, a player, or an HTML5 video element.
- How do I get meeting notes or a summary?
- After transcribing, one click copies a ready-made prompt containing your transcript plus clear instructions to your clipboard. Paste it into whichever AI chat you already use and you get a summary, key points, and action items. Nothing is sent by the copy itself, and there is deliberately no built-in cloud summary, because that would mean sending your transcript to a server.
- What do the Quick refine and Refine with your AI buttons do?
- Both correct words the speech model mis-heard, such as names, places and trade terms that came out as similar everyday words. Quick refine uses the small AI built into Chrome, on your computer: free, private and one click, but it catches only some. Refine with your AI gives you a prompt to paste into ChatGPT, Claude or Gemini, which corrects words using the context of the whole conversation, so it can tell that in a call about a sewer repair “hand date” must be “hand dig”. That context is why it catches the most. Either way the page lists every change so you can keep, edit or reject each one before anything is applied.
- Why doesn't MeetingScribe correct the transcript with its own AI service?
- Because it would break the promise this site is built on: nothing you record leaves your device. MeetingScribe has no server that receives your audio or your transcript, and sending transcripts to an AI service from our side would change that for everyone. Quick refine runs on your own computer, and Refine with your AI happens in a chat you choose and paste into yourself, so what you share stays your decision.
For the full pipeline, every network request the page makes, and two ways to verify the privacy claim yourself, read how it works. If something is wrong or missing, tell us.
Guides worth reading first
Practical writing on the questions that come up before and after a transcription.