AI speech to text, on your device
Record a voice or open an audio file and turn the speech into editable text with Whisper AI, right in this browser. Your audio stays on your device.
- Transcribed on your device
- 99 languages
- Word, PDF, SRT, VTT
- Free to use
This recording
Saved to your account
- Source
- Length
- Windows
- Engine
- Checked when the model starts
- Language
- Not known until it is transcribed
This transcript came from your account; the audio stays wherever you keep it.
Changing a setting transcribes again and replaces your edits.
Long recordings run best on a computer: a phone is slower and can run out of memory.
Keeping stores this recording in this browser until you delete it. Anyone using this device can play it.
The time remaining appears after the first 30-second window.
Marked passages
Marks become an index at the front of the PDF, with the page and line each one landed on.
Subtitles No transcript yet
Document Numbered lines · a timestamp on every line · US Letter
Saving keeps the text, times, speaker labels and marks in your account. Your audio is not uploaded.
Signing in reloads the page, then saves this transcript. Open the recording again afterwards unless it is kept here.
Kept on this device
Nothing is kept on this device. A recording is kept only when you press “Keep on this device”.
These stay in this browser until you delete them, and anyone using this device can play them.
Private speech to text online
Quiet Transcript uses Whisper, an AI speech-recognition model, to turn spoken audio into a timestamped draft on your computer or phone. It understands ninety-nine languages, prints the language it detected, and lets you change it before you rely on the result.
For a long recording, transcribe the first five minutes, check the draft against playback, then run the whole file.
From a spoken recording to editable text
Start by deciding where the audio comes from. You can press Record and speak, or press Open a file and choose a voice memo, an interview or a lecture you already have. Either way the sound is read on your device and nowhere else.
Press Transcribe on this device. The first run downloads the speech model, so the first words take longer to appear than they will on later runs, because your browser keeps the model afterwards.
While the work runs, the progress bar counts real steps: which window the model is on and how much audio that covers. It also shows a time estimate, measured on your device once the first window is done. The result card then names the engine that did the job.
When the transcript appears, press Play and the text follows the recording. Press the time at the start of a line to jump to that moment, and correct the words in place.
An interview, a lecture and a meeting are different jobs
An interview has at least two voices. The tool does not separate speakers, so every label is one you type. Type the name on one line, then use the option that carries a label down, and the lines after it take that name until you type another.
A lecture is one voice for a long time. Run the first five minutes before anything else. If the language and the wording look right, run the whole file. A long recording is cut into 30-second windows that overlap by five seconds, and a phrase that falls on a cut is written once.
A meeting recorded on a laptop is often a video file. Open it as it is. Only its audio track is read, and the picture plays no part in the result.
A voice memo is the simple case: short, one speaker, nothing to set up. Open it, transcribe it, and fix the few lines that came out wrong.
Choosing the language and the model
The language is worked out from the first 30 seconds and printed beside the transcript. A wrong guess does not look like an error. It looks like a fluent transcript in the wrong language, so read the language shown before you trust any of the text.
If the language shown is wrong, choose the right one and the recording is transcribed again with it. English is strongest. Other languages are usable where the model has seen a lot of them and rougher where it has not.
Two models are offered. The small one, whisper-tiny, comes with the first download and is right for most English. The larger one is noticeably better outside English, downloads an extra 80 MB the first time you choose it, and runs slower, especially on a phone.
The tool transcribes and never translates. Speech in Russian comes back as Russian text, in the same language it was spoken.
After the transcript appears
Nothing is kept unless you ask. Reload the page and an unsaved transcript is gone, so download it or save it to your account before you leave.
You can download SubRip (.srt), WebVTT (.vtt), plain text (.txt), Markdown (.md), a numbered PDF or a Word file. Each one is written from what is on screen, so your corrections and speaker labels are in it.
The transcript is a draft. It is not certified, sworn or verbatim. Check it against the recording before you quote it or file it.
More questions about speech to text here
What if my audio file will not open?
Which formats open depends on your browser. The tool asks the browser as the page loads and lists the answer under Which files open here? MP3, WAV and M4A open almost everywhere. If a file cannot be decoded, convert it to MP3 or WAV and open that.
Why was my second transcript faster to start?
The first run downloads the model and its runtime from this site, and your browser keeps them. A private or incognito window forgets them when it closes, so use a normal window if you want the saving.
Is a phone as quick as a laptop?
No. A phone transcribes on the processor, several times slower than a laptop, and a long recording can exhaust its memory. Keep the screen on and the tab in front, and use the small model for long files.
Do I need an account to use this?
No. Transcribing never needs one. An optional Google sign-in saves transcripts and lets you share read-only links that expire after 7, 30 or 90 days. Saving sends the text, times and labels and never the audio.
Does it guess who is speaking?
No. There is no automatic speaker separation. Every label in the transcript is one you typed, so a line never carries a name the tool invented.
How it works, where your data goes and a cloud comparison are on the home page and in the guide.