How to transcribe a meeting, whatever it was recorded on

To transcribe a meeting, first work out what recorded it. If it was Zoom, Teams or Google Meet, the hard part is finding and downloading the file, and each of those has its own guide below. If it is a plain recording, a phone left on the table or a dictaphone or a room system, there is no platform to help you and no built-in transcript to fall back on. That case is the rest of this page, along with what to keep once the actions are written down.

Start with where the recording came from

Zoom, Teams and Meet each have their own file hunt, their own permissions and their own built-in transcript feature. Restating any of that here would only go stale, so the table points at the guide that covers it.

Recorded on	Read	What that guide settles
Zoom	Transcribe a Zoom recording locally	Which file in the recording folder to use, and how to get a cloud recording down first
Microsoft Teams	Transcribe a Teams recording without admin rights	Where the file lands, who is allowed to download it, and what to do when transcription is locked
Google Meet	Transcribe a Google Meet recording	Where Drive keeps it, who gets access automatically, and what the built-in transcript covers
A phone, a dictaphone, a room system or a handheld recorder	Keep reading	Everything below

The plain recording is the case the platform guides miss

A plain recording is a meeting captured by a device in the room rather than by software in a call. Somebody put a phone face down on the table, or the office keeps a handheld recorder in a drawer, or the room system wrote a file to a share. What follows from that is why the platform guides do not transfer.

There is no per-speaker track. A call platform can sometimes write one audio file per participant, because each person arrives on a separate connection. One microphone in a room hears the room. Everyone is mixed into a single waveform before the file is saved, and no tool pulls them apart afterwards.

There is no built-in transcript. Nothing was running except a recorder, so there is no text file next to the audio and no menu that makes one after the fact. Whatever text you get, you make yourself.

There is no permission trail either. On a platform, somebody pressed a button, other people saw an indicator, and there is a record of who recorded what. A phone on a table records whether or not anyone noticed, which puts the consent and retention questions on you rather than on an admin setting.

The file itself is straightforward. A phone recorder app gives you an M4A, a dictaphone or room system often writes a WAV, and a browser recorder writes WebM. Move it to a desktop first, because the transcription step wants a real computer. For the wider picture, the three routes from audio to text lays them out.

What you do before the meeting decides whether the recording is usable

Recording quality cannot be fixed later, and a room recording is far more fragile than a call. Five minutes of preparation beats any amount of cleanup.

Put the recorder near the quiet end of the table rather than the middle. Distance costs you more than anything else.
Put it on the hard table, not on a coat, a notebook or a bag. Soft surfaces swallow the high frequencies that make consonants legible.
Record on one device. Three phones give you three mediocre recordings and a merging problem, not a backup.
Turn off what hums. Air conditioning, a projector fan and a laptop under load sit underneath the voices for the whole meeting.
Say the date, the meeting name and who is present before the agenda starts. That sentence becomes your transcript’s header.
Ask people to say their name the first time they speak. That gives you anchors in a transcript with no labels.

Tell the room it is being recorded, and say what for. That is good manners everywhere and a legal requirement in some places. Nine ways to record audio that transcribes well goes deeper on the acoustics, including the short test recording that catches all of this before the meeting starts.

A meeting transcript is raw material, not the minutes

This is where most people waste an afternoon. A full transcript of a fifty-minute meeting is thousands of words of half-finished sentences, and almost nobody will read it. The decisions and the actions are the deliverable.

Treat the transcript as something to search rather than something to publish. Hunt for the language commitments hide in: “I’ll”, “we’ll”, “let’s”, “by Friday”, “can you”. Pull each hit into a list with an owner and a date, and you have the minutes in fifteen minutes. Keep verbatim text only where the exact wording matters, such as a quoted number or a caveat attached to a promise.

The missing speaker labels bite here, and harder in a room recording than in a call. You get one continuous stream of words with nothing saying who said them. For two people you can follow it from context; for a six-person planning meeting you will be scrubbing through the audio to settle who promised what. What automatic transcription does with two speakers explains why attribution is a separate problem from recognising words. The practical answer is to write names into your notes while the meeting is fresh.

Before anything gets circulated, read it. Seven transcript mistakes to fix before you publish is the checklist we use on our own.

Transcribe a meeting recording with no platform behind it

Once the file is on a desktop, the meeting is just audio. Our transcriber runs OpenAI’s open-source Whisper model on your own graphics card through WebGPU, so the file is read from disk and nothing is uploaded.

Open the site in desktop Chrome or Edge. It needs a working WebGPU adapter, so Firefox, Safari and phones will not run it yet.
Drop the recording on the page. Audio and video both work; only the audio track is read.
The first run downloads about 200 MB of model files, once, and the browser caches them.
Wait. In our test, 71 seconds of audio took 41 to 50 seconds on a desktop with a graphics card, including the model load. A thin laptop takes roughly as long as the recording.
Export TXT for notes, or SRT and VTT if the meeting was filmed. The three exports compared covers the difference.

The limits, plainly. It is English only for now. It uses the base size of the model, chosen to run on ordinary hardware rather than for top accuracy, and names, jargon, strong accents and a bad room are its weak spots, all of which meetings supply. It does not label speakers, which for meetings is the limit that matters most. No account, no length cap, no upload.

Set the delete date before you press record

A meeting recording is personal data about everyone audible in it, and so is the transcript. Settle the storage question early rather than letting files pile up in a downloads folder.

The GDPR states the principle in Article 5(1)(e): personal data must be “kept in a form which permits identification of data subjects for no longer than is necessary for the purposes for which the personal data are processed”. If the purpose was accurate actions, the audio has done its job once the actions are agreed. Article 13(2)(a) expects you to tell people “the period for which the personal data will be stored, or if that is not possible, the criteria used to determine that period”, so you need an answer anyway.

A workable default for an ordinary internal meeting: delete the audio once the notes are circulated and undisputed, keep the transcript while the actions are open, and keep the notes as long as any other record. Decide that once, for every meeting, rather than file by file.

Keeping the whole chain on one machine makes deletion something you can actually carry out, because there is no vendor copy to chase. Personal data in meeting transcripts works through the articles in detail. This is general information about what the regulation says, not legal advice for your situation.

Frequently asked questions

Can I transcribe a meeting recording made on a phone? Yes. Move the file to a desktop first, since transcription needs a real computer with a graphics card. A phone recording is normally an M4A, which works without conversion.

Will the transcript say who was speaking? Not from our tool, and not from any transcript of a single room microphone. You get one continuous stream of text, so add names yourself while the meeting is fresh, using each person’s introduction as an anchor.

How long should I keep a meeting recording? Long enough to write accurate notes, and generally no longer. Article 5(1)(e) ties retention to the purpose, so once the actions are written and agreed, the audio has no purpose left.

Do I need to tell people I am recording? Say so regardless. Rules on recording conversations vary by country, and the GDPR separately expects people to be told what the processing is for and how long you keep it.

Sources, checked 15 September 2026

This post makes no claims about Zoom, Teams or Meet. Every platform detail lives in the linked guide for that platform, which cites the vendor’s own documentation. That leaves two sources.

https://gdpr-info.eu/art-5-gdpr/ : Article 5, principles of processing, including storage limitation at 5(1)(e)
https://gdpr-info.eu/art-13-gdpr/ : Article 13, information to be provided, including the retention period at 13(2)(a)
Our own figures for speed, model size, supported browsers and exports come from our tests of the transcriber on this site