Subtitle Generator
Open the subtitle generator

Runs in your browser

Subtitle generator for video and audio

Add a video or a recording and get timed subtitles you can check over the picture, fix, and download as SRT or VTT. An open-source speech recognition model does the work on your device, so the file is never uploaded. Free, and there is no account.

Read on your device. Never uploaded.

Model

The larger model is a bigger download and needs more memory. How the two compare will be published here once they are measured on this engine.

Step by step

How to make an SRT file from a video

  1. 1

    Add your video or audio file

    Drop it on the box above, or click the box to choose it. The page reads the file on your device.

  2. 2

    Pick the language and the model

    Leave the spoken language on Detect automatically, or choose it from the list. The smaller model is the default and the lighter download, about 60 MB; the larger one is about 264 MB and needs more memory.

  3. 3

    Wait for the first cues

    Your first file downloads the model once. Then the subtitles appear a few minutes of audio at a time, and the status line shows how far it has got, with an estimate of the time left measured on your device.

  4. 4

    Check them over the picture

    Play the video and each cue shows over it at its time (with audio, under the player). Fix a word in the text box, split a long cue at the cursor, merge two short ones, or move a start or an end by a tenth of a second.

  5. 5

    Download SRT or VTT

    Both files are named after your video and hold the cues as you edited them. Not sure which one you need? The guide below compares them.

The result of a real run: the cue over the paused video, the line that names the language, the model, and the engine build, and the cue with its timing buttons.

The subtitle tool after a run: the preview with the cue Concorde returned to its place amidst the tents. over a paused test video, the line Detected language: English, smaller model, WebAssembly on several threads, the cue editor with Cue 1 from 00:00:00.570 to 00:00:03.350, and the Download SRT, Download VTT, and Start over buttons.
Test clip: a LibriSpeech sentence (CC BY 4.0) over generated color bars.

What each control does, with screenshots · SRT or VTT: which file you need

What the subtitle file holds

  • Cues of at most two lines of about 42 characters, each on screen for at most 7 seconds, and for at least 1 second when the gap to the next cue allows.
  • Cuts between words: a cue ends at the end of a sentence first, then at a comma or a pause, then where the parts come out even; inside a cue, the two lines break after punctuation when that keeps them close to even.
  • A start and an end for every cue, to the millisecond, taken from the times the model gives each word.
  • Your cues and nothing else: no watermark and no credit line.

What it does not do

  • Translate: the subtitles are in the language that is spoken.
  • Name the speakers or describe sounds, as full closed captions do.
  • Burn the subtitles into the video: you get a file to add in your editor, player, or platform.
  • Open an SRT or VTT file you already have: it edits the subtitles it makes.
  • Take a link or a microphone: it works on a file on your device.

How the subtitles are made in your browser

Privacy by design

Your file stays on your device

No upload

The page reads your file with your browser's File API and makes the subtitles in the tab. The site is static files with no upload endpoint, and its security policy lets the page connect only to this site.

No account, nothing to install

Open the page and add a file. There is nothing to sign up for, no cookies, no analytics, and no ads.

Check it yourself

Your browser's developer tools list every request the page makes. You will see the page, the engine, and the model come in, and nothing of your file go out.

Check that nothing is uploaded, step by step

FAQ

Questions about the subtitle generator

Is Subtitle Generator free?

Yes. There is nothing to pay and no account to create. The speech recognition runs on your own device, so the site has no per-file cost to pass on to you.

Is my video uploaded?

No. Your browser reads the file and the model makes the subtitles inside the tab. Besides the page itself, the only downloads are the engine, the model, and a small voice detector, all from this site, and the page sends nothing back. See how to check that nothing is uploaded.

Which languages can it write?

The language list holds every language the speech recognition model knows, and the page can detect the language for you. How good the subtitles are varies a lot between languages, speakers, and recordings, and we have not published measured results per language yet. Our automated test checks English speech, so detection is checked on English only so far. Read the first minute of your subtitles before you rely on the rest, and if the detected language is wrong, choose it yourself and run the file again.

How accurate are the subtitles, and how close are the times?

We have not measured either on this engine yet, so we give no figure. Expect some wrong words, especially with music, noise, or several people talking at once, and times that can be a little early or late. The editor lets you fix both: type over a word, or move a start or an end by 0.1 second. How the subtitles are made in your browser explains where the times come from.

How long can my file be?

Where your browser can read the file's sound a slice at a time, the page sets no length limit of its own. The longest file we have run so far is 12 minutes. Where the browser cannot, the page decodes the whole file at once, and then it stops at 500 MB or 1 hour and tells you which. Either way your device does all the work, so a long file takes a long time: the status line shows an estimate once it has measured some of it.

Which video and audio formats work?

Our automated test runs WAV, MP4 (H.264 video with AAC sound), and WebM (Opus sound) files in Chromium, the engine of Chrome and Edge. Other files work when your browser can decode their sound. If it cannot, the page says so, and saving the file in another format can help.

Should I download SRT or VTT?

SRT is the older, simpler format that many editors and players open; VTT (WebVTT) is the format a web page's video player reads. The page writes both from the same cues. SRT or VTT: which file you need shows the same subtitles in both, line by line.

Can I edit the subtitles before I download them?

Yes. Change the text of any cue, split a long cue where your cursor is, merge a cue with the next one, and move its start or end by 0.1 second. The preview follows every change, and the downloads write the cues as they are at that moment. What each control does walks through them with screenshots.

Does it add the subtitles to my video?

No. It makes a separate subtitle file, and your video stays as it was. You add the file in your video editor, your player, or your video platform. Subtitles drawn into the picture (burned in) need a video editor.

Can I use the subtitles on YouTube?

YouTube's own help page, Supported subtitle and closed caption files, lists SubRip (.srt) and WebVTT (.vtt). We have not tested an upload to YouTube ourselves yet, so this site does not give the steps.

Is there a watermark or anything added to the subtitles?

No. The files hold your cues and nothing else, and our automated test checks that the downloaded cues equal the ones on the page. One exception: if a run stops before the end of your file, the VTT file says where the subtitles stop, in a comment line that players do not show.

Is it AI?

It uses a speech recognition model, a kind of machine learning model, and that model runs in your browser tab on your device. Nothing is sent to an AI service. How it works names the model and its license.

Curious what happens to your file?

It stays in your browser tab. See what the page downloads, what your browser keeps, and how to check it.

See what happens to your file