Transcribe video to text for free

Where your file goes

Nowhere. The audio is read and transcribed inside your browser, by your device's processor or graphics card. The internet is only used to download the program and the model: the library comes from jsDelivr, parts of the video engine come from unpkg, the text fonts come from Google Fonts and the Whisper model comes from Hugging Face. None of those addresses receive your file, and no Tavea server takes part in the transcription. The site only logs the page visit, with nothing from the file. Want to check? Transcribe a file, turn off your internet and, in the same tab, transcribe another one: it works the same.

Limits, no sugarcoating

• The first run downloads the model. That's about 27 MB for the engine plus 80 MB for the accurate model or 44 MB for the fast one. Audio over 5 minutes, on a computer with a graphics card, uses a larger version of the model: 209 MB. Burning subtitles into an MP4, or opening a file over 250 MB or in a format the browser can't read, downloads another 32 MB for the video engine. All of it stays in your browser cache, so later runs skip the download. On a phone, use Wi-Fi.
• The model is small. It gets names, brands and technical terms wrong, and it does worse with noise, loud music or several people talking at once. Proofread before you publish.
• In stretches with no speech, it can make up a sentence. We filter out the most common ones, but check any silent or music-only parts.
• Phones: the limit here is 200 MB and 20 minutes of audio, so the tab doesn't crash from lack of memory. On a phone with little memory, even that can fail. On a computer the limit is 2 GB and 2 hours.
• Time: it depends on your device. In our test on a desktop computer, with the model already loaded, a 42 s video took 8 s on the processor, and 6 min of audio took 57 s on the graphics card. A weak device can take longer than the audio itself. Keep the tab open until it finishes.

What a Reel transcript is good for

Reading your own video shows you things your ear lets slide: the greeting that delays the topic, the promise that only shows up at second 8, the repeated line. Take the first line of the transcript and run it through the hook analyzer. The text also becomes your post caption, the script for your next video and the base for on-screen subtitles.