How it works
Subtitles are generated in three steps, entirely on this device.
- 01Choose a fileVideo or audio in a common format. It is read on this device, not uploaded.
- 02GenerateWhisper transcribes the speech. The model downloads once and stays in the browser.
- 03Edit and exportCorrect the text, then download SRT, VTT or plain text, or a copy of the video with subtitles.
FAQ
Privacy, accuracy and requirements.
Is my file uploaded?
No. Audio decoding, speech recognition and video export all run in your browser. The only download is the speech model, from Hugging Face.
Which model should I choose?
Accurate uses Whisper large-v3-turbo (about 620 MB) and requires WebGPU. Fast uses Whisper base (about 110 MB), runs on more devices and makes more mistakes, especially in languages other than English.
How long does it take?
On a recent laptop with WebGPU, an hour of speech takes about 15 minutes with Accurate and 5 with Fast. Silence is skipped.
Which languages are supported?
More than 30, including English, Chinese, Japanese, Spanish, Portuguese and German. Choosing the language gives better results than automatic detection.
Can the subtitles be added to the video?
Yes. Burning them in draws the text into the picture and exports an MP4 with the original audio. The preview shows the size and style used.
Does it work offline?
Yes, once the page and the model have loaded. The model stays in the browser until site data is cleared.
Which browsers are supported?
Current versions of Chrome, Edge, Safari and Firefox. The Accurate model requires WebGPU, available in Chrome and Edge on most computers.