Dubash: Auto Captions

Dubash writes captions for videos spoken in any of 99 languages. Pick a video, choose the language, and a few minutes later you have timestamped captions that you can edit, style, burn into the video or save as an SRT file. The captions can be in English, or As spoken (in beta): in the language of the video, written in its own script, so a Telugu video gets Telugu captions.

Everything happens on your phone. After a one-time download of the speech model, your video and your captions never leave the device. There are no accounts and no ads.

Dubash understands 99 languages, including Hindi, Telugu, Tamil, Spanish, Arabic, Chinese and Japanese. It can usually work out which language is being spoken, or you can tell it.

How it works

1. Choose a video from your phone, or share one to Dubash from another app.
2. Choose English or As spoken captions, and Best or Fast. Best gets more words right. Fast is quicker and a smaller download.
3. Wait while Dubash listens. It skips the quiet parts, shows how far along it is and roughly how long is left, and you can cancel at any time.
4. Read through the captions and fix any the model misheard. Then export an MP4 with the captions drawn on, or an SRT file for YouTube or a video editor.

Styling

Change the size, position and colours of all the captions at once. The first time you open the editor, Dubash asks whether you want just captions or some flair. With flair, you can pick a line and give it a look of its own: a size up to Huge, its own place on the frame, an outline, capitals, its own colours or a tilt. Turn on Light up each word as it is spoken, and each word is highlighted as it is said, in the colour you choose. The Stickers tab lays pictures from your phone, coloured shapes and short clips of other videos over yours; drag one into place, then size, turn or recolour it. Choose Just captions and the editor keeps to the caption list and Export.

Unless you move them, captions sit above the area that Instagram's Reels interface covers, so they stay readable once posted.

The speech model is a one-time download of 190 MB (Fast) or 539 MB (Best). Your phone keeps it, so you only wait once, and after that Dubash writes captions without internet. The app does send anonymous usage counts: how often captions are started, finished and exported, never anything about your videos. Apart from those, it sends a feedback message or a star rating only if you choose to.

Dubash is the old Indian word for an interpreter.