How turning a video into text works
It gets all the text out of a video: what people say, and the text shown in the picture, like slides, captions burned into the video, titles and signs. You get one list in time order, read on your own computer, so a recorded meeting or a lecture is never sent anywhere.
Turn a video into text in three steps
- 1Choose your video.
MP4, MOV, WebM or MKV. Leave the choices as they are for everything, or choose only the spoken words or only the text on screen.
- 2Check the start.
The free preview reads the first 30 seconds and shows the first lines.
- 3Download, correct, save again.
You get a .txt file with every line and its time. Correct any line on the page, then save the text again, or the spoken words as subtitles (.srt or .vtt).
What it looks at
- The sound: the speech model listens and writes down what is said, with the time of each sentence.
- The picture: every 2 seconds it takes a still and reads the text on it. Stills that look the same as the last one are skipped.
- Only words it is sure of are kept, so specks and patterns are not read as letters.
- A slide that stays on screen for a while appears once, at the time it first shows.
- In the text file, spoken lines read "[0:12] …" and text on screen "[0:15] On screen: …".
What it can't do
Like any automatic reading, it makes mistakes, so read the text through before you use it.
- Handwriting, very small text and text that moves are read badly or not at all.
- The picture is checked every 2 seconds, so text shown for a shorter time can be missed.
- The language you choose is also used to read the text on screen. Text in another language is read less well.
- "More accurate" understands other languages and noisy sound better, but it is a bigger download the first time (240 MB instead of 77 MB) and slower. It is best on a computer.
- It needs a laptop or desktop computer: the work is too heavy for a phone.
Your video stays with you
The video is read on this device and never sent to a website or a server. After the first time, the speech model and the reader stay in your browser.