Auto captions, on your device.
Free · In your browser · No signup
On a phone this can be slow and may run out of memory. A laptop is more reliable. Keep the clip under 60 seconds.
Edit any line, then download.
Your files never leave your device. This tool runs entirely in your browser: nothing is uploaded, stored or sent to a server.
Whisper speech recognition runs inside your browser. Drop in a clip of up to a minute, get timed captions, edit them and download an SRT. The audio stays on your device.
How to use it
- Choose a clip of up to 60 seconds. If it is longer, only the first minute is transcribed.
- Pick a model. Tiny is fastest and best for clear speech; Base is more accurate and takes longer to load. Leave the language on auto-detect or choose it.
- Press Transcribe. The first run downloads the model, about 40 to 75 MB, which your browser then keeps. Transcription runs on your processor.
- Edit the text if needed, check the caption preview on the video, then download the SRT file and upload it to your editor or ad platform.
Why captions belong on every ad
A large share of short video is watched with the sound off, so captions are often the message. Meta and TikTok let you upload an SRT, and every editing tool imports one. Clear, correctly timed captions also make an ad easier to follow in a noisy place and more accessible to deaf and hard-of-hearing viewers.
Keep each caption to one or two short lines, and keep the text out of the platform's bottom interface zone. Our safe zone checker shows where that is.
What to expect from in-browser Whisper
This is the open-source Whisper model compiled to run in your browser. It is good on clear, close-microphone speech and weaker on heavy accents, music beds and several people talking at once. It will occasionally mishear brand names and ingredient names, so always read through the text before you publish.
It runs on your own device, so speed depends on it: a modern laptop transcribes a 30-second clip in well under a minute with the Tiny model, while an older phone can be slow or run out of memory. We recommend a desktop browser. The model files are fetched once from public model hosting, and your audio is never sent there.
Privacy
Your files never leave your device. This tool runs entirely in your browser: nothing is uploaded, stored or sent to a server. The page loads its code once and does the work on your device. There is no upload, no account and no watermark, and nothing you add is stored on our servers.
Questions
About this tool.
Is my audio uploaded?
No. Audio is decoded and transcribed in your browser. Only the model files are downloaded, from public hosting (jsDelivr and Hugging Face), and your clip is never sent to them or to us.
Why is there a 60 second limit?
Running a speech model in a browser uses a lot of memory and time. Sixty seconds covers most ads and keeps it reliable. Split longer videos into parts.
Does it work on mobile?
It can on a recent phone, but the model download and the processing are heavy. We recommend a laptop or desktop.
Which languages are supported?
Whisper handles many languages. Tiny and Base are weaker on some of them, so check the output carefully for anything other than English.
Is the text always right?
No. Brand names, ingredient names and noisy audio are the usual errors. Edit the text before you download the SRT.
Can it burn the captions into the video?
This page shows a live caption preview over the video and gives you the SRT, which every editor and ad platform can apply. Export of a burned-in file is not included, which keeps the tool light and reliable.