Tesseract OCR
Opens DVD and Blu-ray subtitles that are pictures, not text, by reading the words in them with Tesseract 5.5. Their timings and positions stay as they were, and the text is yours to edit, translate and save in any format. It runs on your own computer's processor, offline, in more than a hundred languages.
Settings Add-ons Tesseract OCR Install
Subtitld downloads it, checks its SHA-256, and keeps it updated.
- Languages
- 108
- Models
- 2
- Add-on download
- 13–22 MB
- RAM, at least
- 256 MB
Needs a newer Subtitld. The add-on is in the catalog, but reading picture subtitles comes with the release after Subtitld 26.10. Subtitld 26.10 can install it and does nothing with it.
What it does
DVDs and Blu-rays carry subtitles as pictures: VobSub (.idx and .sub) and PGS (.sup). Subtitld draws each one out of the file and Tesseract reads its words, so you get ordinary subtitles to edit.
Each subtitle keeps the start and end of its picture. Subtitles shown at the top of the screen stay at the top.
Picture subtitle tracks in an MKV or other video are read the same way, from the Import panel.
No graphics card and no internet once a language is there. A film's thousand or so subtitles take about a minute.
Opening a file
Open the .idx, .sub or .sup file as you would any subtitle file. Subtitld sees that its subtitles are pictures and asks which language they are in: Tesseract reads better when it knows. A VobSub's .idx usually names its language, and Subtitld picks it for you; a .sup never does, so check it. Then click Read.
A bar shows the reading as it goes, and closing the dialog stops it. Subtitles Tesseract isn't sure of are read a second time with their colours the other way round, which helps on discs that draw dark letters with a light outline. When it's done, the subtitles open in the editor like any others.
Saving writes a new file, as .usfx, or in any format from Export. The .idx, .sub or .sup you opened is never changed.
Without the add-on, Subtitld says the subtitles are pictures and that this add-on reads them.
Subtitles inside a video
In the Import panel, choose Subtitles in the video. With the add-on installed, the video's Blu-ray (PGS) and DVD (VobSub) tracks can be chosen too: they show as pictures, read by Tesseract OCR. Text tracks are still preferred when the video has both.
DVB subtitles (from TV recordings) and XSUB (from some DivX files) can't be read yet.
Models
Tesseract reads with one model per language, from the Tesseract project's tessdata_fast set: the quick models, 1 to 12 MB each. English is inside the add-on. Every other language downloads into Subtitld's shared model cache the first time a subtitle in it is read, from a pinned version of tessdata_fast, each file checked against its fingerprint. Updating or removing the add-on keeps them.
| Model | Download | Good for |
|---|---|---|
| English (tessdata_fast) Default eng | English, included in the add-on | |
| Other languages (tessdata_fast, one file per language) tessdata_fast | Every other language, one file each, downloaded the first time a subtitle needs it |
Options
In Settings › Add-ons › Tesseract OCR › Configure.
Tesseract reads quotation marks as curly ones. On, they become straight, as most subtitle files write them; turn it off to keep them curly.
Languages
More than a hundred, Latin, Cyrillic, Greek, Arabic, Hebrew, Indian, Chinese, Japanese and Korean scripts included. Chinese has two models, simplified and traditional: Subtitld picks traditional for zh-TW and zh-HK. Serbian, Azerbaijani and Uzbek, written in two scripts, are read in their usual one unless the language names the other.
afamarasazbebgbnbobrbscacebchrcocscydadedvdzeleneoeseteufafifilfofrfygagdglguhehihrhthuhyidisitiujajvkakkkmknkokukylalbloltlvmimkmlmnmrmsmtmynenlnoocorpaplpsptqurorusasdsiskslsqsrsusvswsyrtatetgthtitotrttugukuruzviyiyozh
Tesseract reads clean, plain subtitles very well. It does less well on small, bold DVD lettering, where a word or two may come out wrong ("net" for "not"): read the subtitles through before you use them.
Platforms
| System | Download | Tested |
|---|---|---|
| Linux x86-64 | 21.6 MB | Read VobSub and PGS files in Subtitld, on a real machine and in CI |
| macOS (Apple silicon) | 18.0 MB | Read test subtitles in CI (Apple Silicon) |
| Windows x86-64 | 12.9 MB | Read test subtitles in CI |
Linux x86-64 (glibc 2.35 or newer, as on Ubuntu 22.04, Debian 12 and RHEL 9), macOS on Apple Silicon and Windows x86-64. Tesseract is built from its own sources for each of them, with nothing to install beside it. On each one, CI installs the add-on as Subtitld does and reads test subtitles with it, one of them in a language it downloads.
Troubleshooting
networkA language model couldn't be downloaded. Check your connection and open the file again.unsupported_languageTesseract has no model for the language chosen. Pick the language again in the dialog; for a language it doesn't know, a related one in the same script often reads well.model_missingA file the add-on needs is missing. Reinstall it from the Add-ons panel.oomThe computer ran out of memory. Close other programs and read the subtitles again.bad_paramsSubtitld asked for something the add-on doesn't understand. Update both.internalSomething failed inside the add-on; the message has the details.
License
Fine for commercial work: the add-on, Tesseract and the tessdata_fast models are Apache-2.0, and Leptonica is BSD. The text you read is yours, but the subtitles on a disc belong to whoever made them: use them as their licence lets you.
Versions
- 26.11.1
The first release. Subtitld decodes DVD VobSub (.idx/.sub) and Blu-ray PGS (.sup) subtitles, and the picture tracks of videos, into one picture per subtitle; this add-on reads the text in each with Tesseract 5.5.4 (LSTM engine, tessdata_fast models). It reads several pictures at once, reports progress after each, and stops when cancelled.
English is included. Any of 100 other languages downloads the first time a subtitle needs it (1 to 12 MB each), checked against a pinned sha256. Straight quotes in place of curly ones are a setting.
Runs on the CPU, on Linux x86-64, Windows x86-64 and macOS on Apple Silicon. Needs a Subtitld that knows the ocr.recognize task.