Make subtitles for a video or a recording

An interview, a lecture, a client call, a product demo: the recordings that most need subtitles are often the ones that should not be sent to a transcription service. Here the speech model runs inside this tab, so the file never leaves your computer.

Loading the tool…

How it works

  1. Drop the video or audio file. Leave the language on automatic, or choose it if the automatic guess is wrong.
  2. Press “Create subtitles”. The first time, the speech model is downloaded once (240 or 540 MB, depending on your computer); then the recording is transcribed part by part.
  3. Correct the text and the times while the video plays them back, then download .srt or .vtt, or make an MP4 with the subtitles burned in.

Why nothing is uploaded

Every operation on this page is done by code running inside your browser tab, using the same engine that renders web pages. The file is read from disk into your tab’s memory, transformed there, and written back out as a download. It is never sent anywhere — not to us, not to a third party.

Verify it yourself

  1. Open your browser’s developer tools (F12) and select the Network tab.
  2. Load your file and run the tool.
  3. The only requests you will see fetch the tool’s own code — and, for a few heavy tools, their open-source engine (also from this site) or an AI model from Hugging Face — plus one small page-view ping to loreatec.jp (page address and title, nothing more). None of them carry your file.

Proof it stays local →

Frequently asked questions

How accurate is it?

With clear speech and one person talking at a time, most words come out right, but treat the result as a draft. Accuracy drops with background music, people talking over each other, strong accents, jargon and poor microphones. Check names, numbers and technical terms, and read it once against the video. During music or silence the model sometimes writes a line nobody said: delete it. Times can be off by a fraction of a second.

Which model is used, and why is the download so large?

OpenAI Whisper (MIT licence), in the builds published on Hugging Face. When the browser offers WebGPU, large-v3 turbo (about 540 MB) runs on the graphics chip and times every word. Elsewhere Whisper small (about 240 MB) runs on the processor: slower, less accurate, and timed by phrase rather than by word. Either one is downloaded once and kept in the browser; the local AI page of this site uses the same files, so they are not downloaded twice.

Is my recording uploaded?

No. The sound is decoded, transcribed and turned into subtitles inside this browser tab. The only downloads are the model itself and, for some formats, the video engine served by this site. You can check it: once the model is loaded, the tool keeps working in flight mode.

How are the lines split?

At most two lines per subtitle, about 42 characters per line (16 in Japanese and Chinese), and between one and seven seconds on screen. Lines break at the ends of sentences, at pauses and at commas, and not after an article or a preposition when that can be avoided. Every subtitle can be edited, joined with the next one or deleted, and its times changed by hand.

SRT, VTT or burned in?

SRT is accepted almost everywhere: YouTube, Vimeo, video editors, VLC. VTT is the format of web pages (the <track> element) and of some platforms. Burning in draws the text onto the picture, for places that cannot load a subtitle file; it re-encodes the whole video in the browser as an H.264 MP4, which takes a while, and the subtitles can no longer be switched off.

How long does it take?

It depends on the computer. On an ordinary laptop with built-in graphics it took 1 to 1.5 times the length of the recording; on the processor alone, about twice the length. A dedicated graphics card is faster. The recording is transcribed in 30-second parts, the progress bar shows which part is being worked on, and you can stop at any time.