Discover tools
All tools

Transcribe audio

Turn speech in audio or video into text and subtitles with Whisper, running on your device. Free, with no account or uploads.

Drop audio or video to transcribe

What is said, written out with times by Whisper running on your device: search it, play along, and save it as text or subtitles. Nothing is uploaded. You can also paste with ⌘V or Ctrl+V.

  • MP3
  • M4A
  • Opus
  • FLAC
  • WAV
  • AIFF
  • MP4
  • MOV
  • WebM

Stays on your device · up to 50 files at once

This page is the Audio editor, opened for transcribing.

What you can do

  • Trim, fade and change the volumeTo the millisecond, on a waveform you can zoom; every edit can be undone, and your file is never changed.
  • Every common formatMP3, M4A (AAC), Opus, FLAC, WAV and iPhone ringtones, from those and from AIFF, Ogg and the sound of videos.
  • Loudness to a standardMeasured as ITU-R BS.1770 does, normalized to −14 LUFS or another target, with a true-peak limiter so nothing clips.
  • Sound out of videoKept exactly as it is, without being encoded again, whenever the format allows; or converted to the format you choose.
  • Checked before you downloadEvery file written is opened again and listened to: its length, rate and channels are held against what you asked.
  • Many files at onceTags and cover art kept, gapless MP3s, a size limit if you need one, and everything in one ZIP.
  • Nothing is uploadedYour files are read and written in this browser. No account, no limits.

A little help along the way

Follow the steps below. Every option is explained underneath.

  1. Add your filesDrop them, choose them or paste them: MP3, M4A, WAV, FLAC, Opus or videos.
  2. Choose the modelBase is quick; Small is more accurate and slower. Either is downloaded once, then kept.
  3. TranscribeThe text comes in as Whisper hears it. Stop at any time and keep what was heard.
  4. Read, search and savePress a time to listen from there, find words, and save it as text or subtitles.
Every option explained
Whisper
OpenAI's speech recognition model, run in your browser with Transformers.js and ONNX Runtime. The first time, the model and its engine are downloaded from this site (Base about 107 MB, Small about 280 MB) and then kept for working offline; the sound never leaves your device. Base takes about a third of the sound's length on a laptop and is good for clear speech; Small is slower and better with accents, noise and other languages.
Language
Whisper finds the language by itself, or you can say which it is, which helps with short or mixed speech. Translate into English writes what is said in English instead: a quick translation, not a careful one.
The transcript
Written in parts with their times, as Whisper hears them, 30 seconds at a time; each next window starts where the last whole sentence ended, so nothing is cut in two. Press a time to listen from there, find words in it, copy it, or save it as text, text with times, SRT or WebVTT subtitles. Whisper can mishear names, quiet words and speech over noise, and on silence or music it can write words that were not said: read it through before you rely on it.
Same as the file
Keeps the file's own sound as it is, without decoding and encoding it again: AAC goes into an M4A, MP3 stays MP3, Opus and Vorbis go into Ogg, FLAC stays FLAC and plain samples become a WAV. Nothing is lost and it takes a moment. When the rate, the channels or a size limit you chose mean the sound must change, it is encoded again in its own kind; sound that cannot be kept (such as AC-3 from a film) becomes AAC, and the page says so.
MP3
Plays everywhere. Written with LAME at a constant bitrate, with the header that tells players the encoder's delay and padding, so the file is exactly as long as the sound. Below 32 kHz an MP3 holds at most 160 kbit/s.
M4A (AAC) and M4R
Smaller than MP3 for the same quality, and played by every phone and computer. Written with your browser's own AAC encoder, or one that comes with the page when it has none. The encoder's priming is measured and an edit list written, so the sound starts and ends where it should. M4R is the same for an iPhone ringtone, which can be at most 40 seconds long (an alert tone 30).
Opus
The best quality for its size, voice above all. It is always written at 48 kHz. Browsers, Android, VLC and recent Apple devices play it.
FLAC and WAV
Lossless: every sample is kept. FLAC is about half the size of WAV. Choose 16-bit (CD quality), 24-bit, or for WAV 32-bit floating point; FLAC holds at most 24 bits.
Quality
Small, Balanced or High set the bitrate for each format (MP3 96, 192 or 320 kbit/s; AAC 96, 160 or 256; Opus 48, 96 or 160), less for mono. Custom lets you pick the bitrate. These numbers are KeepSimp's own choices.
Sample rate and channels
Keep the file's own, or choose one. When a format cannot hold the rate asked, the nearest it can is used and the page says so. Lossy formats here are mono or stereo, so surround sound is mixed down to stereo for them.
Tags and cover art
Title, artist, album, year, genre and the cover picture go into the new file where its format holds them. Switch it off to write none.
File size limit
MP3, M4A and Opus are written at the highest bitrate that keeps each file under the limit, with room for the container; the page says what it chose. Lossless formats cannot be made to fit a size.
The check
Every file written is opened again and decoded from start to end. Its length, sample rate and channels are held against what was asked, an MP3's size against its bitrate, and a file that came out silent is caught. If something is not as asked, the page says what.
Files the browser cannot read
Some browsers' decoders lack a kind of sound (Safari's lacks FLAC in some places). The page then decodes it with the browser's own player, which plays more kinds, and goes on. AIFF is read by KeepSimp itself. A file nothing here can decode says so.
Privacy
Files are read and written in this browser and never uploaded. There is no account and no limit on how many you convert.
Read the step-by-step guide