Audio to text, transcribed free in your browser

Turn audio to text from a meeting, an interview, a lecture or a voice memo. Open the MP3, WAV or M4A here, Whisper writes down what was said on your own computer, and you keep a plain .txt transcript. The recording is not uploaded, and you can try it without an account.

Free. Works with MP3, WAV, M4A, AAC, OGG, FLAC, Opus and WebA files.

Transcribe audio to text in five steps

There is nothing to install. The transcriber is the same app that sits on the OpenSubs home page; it simply accepts sound files as well as videos.

  1. Go to the app and drop your recording onto the page, or press Choose a file and find it in your downloads or voice-memo folder.
  2. Only need part of it? Under Clip & size, set where the transcript should begin and end, so the small talk before a meeting starts is left out.
  3. Tell it which language is spoken, or let it detect the language, and press Generate from the audio. Give the first run a minute: it fetches the speech model once and your browser keeps it for next time.
  4. Play the recording and read the lines as they go by. Fix names, jargon and numbers by typing straight into a line.
  5. On the Export tab, press .txt. You now have the words of the recording as a text file you can open in any editor.

What the .txt transcript looks like

The text file holds only the spoken words. Short lines are joined into running sentences, and whenever the speakers pause for two seconds or longer, a new paragraph begins. A team call might come out like this:

Okay, I think everyone is here. Let's start with the launch date, because the printer needs the final files by Thursday.

I spoke to them this morning. Thursday is fine if we send the cover tonight, otherwise it moves to Monday.

Then let's send it tonight. Priya, can you check the spine text one more time before six?

Notice what is missing: there are no line numbers, no timecodes and no names in front of each turn.

Plain text, or text with timestamps

A transcript for reading and a transcript for finding moments are different things, so the Export tab offers both. Pick by what you will do next:

You want toSave
Paste the words into notes, an email or a document.txt, the words alone
Quote someone and point to the minute they said it.srt or .vtt, every line with its start and end time
Caption a podcast clip or a talk you will publish.srt; the audio to SRT guide covers that route
Keep a styled subtitle file for a video editor.ass

All four exports come from the same edited lines, so a correction you type once appears in every file you save.

Recordings people turn into text

Meetings and calls

Recorded a call with an app that saves M4A or WAV? Transcribe it afterwards to pull out decisions and action items without listening again.

Interviews

Journalists and researchers get a first draft of the conversation to check against the tape, instead of typing every answer by hand.

Voice memos

Ideas dictated into a phone while walking become text you can search, sort and copy into a to-do list.

Lectures and classes

A recorded lecture becomes study notes to skim, highlight and search before an exam.

Podcast episodes

A text version of an episode is a starting point for show notes, a blog post or a list of quotes.

Voice messages

A long voice note saved as an OGG or Opus file can be read in a minute instead of played in full.

Languages, and turning the text into another one

Whisper understands speech in many languages. Choosing the language yourself usually gives a better result than leaving it to guess. If a recording goes back and forth between two languages, that is handled too: the app re-checks the spoken language every four seconds, and both languages end up in the same transcript.

Need the words in a different language? After the transcript is ready, use Translate. In desktop Chrome 138 or newer the built-in translator runs on your computer at no cost; you can also use your own API key or paid OpenSubs credits. If you keep the original visible alongside the translation, the .txt keeps each passage in both languages, one under the other. How translation works.

Getting a cleaner transcript

The speech model comes in three sizes. Tiny (about 40 MB) only knows English. Base (about 80 MB) is the everyday choice. Small (about 250 MB) makes fewer mistakes but needs roughly double the time. Whatever you pick, expect to correct some words: real rooms have echo, crosstalk and accents, and no model gets every name right.

  • Record close to the speaker if you can. A phone on the table in a quiet room beats a laptop microphone across a busy office.
  • Switch to Small when people talk over each other or speak quickly.
  • Spell out unusual names once in the editor; it takes seconds and the export follows.

Long recordings and how fast it runs

There is no cap on file size or length, and no queue to wait in. Speed depends on your own hardware: on a computer with a capable graphics chip the browser uses it through WebGPU, and otherwise it falls back to the processor, which is slower. A one-hour lecture on an older laptop can take a good while, so leave the tab open and let it finish. For files of several hours, the long recordings guide explains what to expect and when the desktop app is the better fit.

Where your recording goes

With the default setting, transcription happens inside the browser tab, so a confidential meeting or interview stays on your computer. What does travel over the network is the app itself and, on the first run, the speech model. You can watch this in your browser’s developer tools under Network. There is also an optional cloud transcription mode that you would have to switch on and give your own key for; only then is the trimmed audio sent, to the service you chose. The Whisper in the browser page explains how the local model runs.

What this audio transcriber does not do

  • It does not label speakers. The transcript has no “Speaker 1” or names; add them yourself while you edit if you need them.
  • It does not summarise or write meeting minutes. You get the words that were said, not a shortened version.
  • It does not record from your microphone. Start from a recording that is already saved as a file.
  • It handles one recording at a time; there is no batch upload of a whole folder.
  • It cannot run in iOS 17 Safari. iOS 26 works.

Starting from a video rather than a sound file? See transcribe video to text.

Questions

How do I convert audio to text for free?

Drop the MP3, WAV, M4A or other audio file onto opensubs.app, press Generate from the audio, correct any wrong words, then press .txt on the Export tab. It costs nothing and you do not need an account.

Is my audio uploaded to a server?

No, not on the default setting. Whisper runs inside your browser tab and the recording is read from your own disk. The only download is the speech model, the first time.

Can I get a transcript with timestamps?

Yes. The .txt has no times, but the same transcript saved as .srt or .vtt gives every line its start and end time.

Does it tell different speakers apart?

No. The transcript contains the words only, without speaker labels. You can type names into the lines while editing.

Is there a length limit for the audio?

No. There is no file-size or length limit. Long recordings simply take longer, because the work is done by your own computer.

Can I transcribe audio on my phone?

It depends on the browser. On iPhone, Safari on iOS 26 can run the speech model and iOS 17 cannot. Long recordings go faster on a computer.