PP-OCRv5 (offline OCR)
Opens the subtitles of DVDs and Blu-rays, which are pictures, not text, and turns them into subtitles you can edit, with PaddleOCR's PP-OCRv5. It reads them on your own computer's processor, with no graphics card and no account, in about a minute or two for a film, and nothing leaves your machine.
Settings Add-ons PP-OCRv5 (offline OCR) Install
Subtitld downloads it, checks its SHA-256, and keeps it updated.
- Languages
- 77
- Models
- 12
- Add-on download
- 127–150 MB
- RAM, at least
- 512 MB
What it does
DVD subtitles (VobSub, an .idx with its .sub) and Blu-ray subtitles (PGS, a .sup) are drawn as pictures. PP-OCRv5 reads every picture as text, so you can edit, retime and translate them, and save them as SRT or any other format.
Open the subtitle file itself, or read a picture track straight from an MKV, MP4 or TS.
Each subtitle keeps its own start and end, and a subtitle drawn at the top of the picture stays at the top.
All twelve models come inside the add-on. It runs on the processor, about 25 ms a line, and nothing leaves your computer.
Before you install
The add-on works only with a version of Subtitld newer than 26.10, the first that can open picture subtitles. Older versions list it but have nothing that uses it.
- Disk: about 260 MB once installed, all models included.
- Memory: about 512 MB free while it reads.
- Processor: any 64-bit Intel or AMD processor, or Apple Silicon. No graphics card is used.
Opening picture subtitles
A subtitle file. Open a .idx, .sub or .sup as you would any subtitle file. Subtitld asks for the Language of the subtitles, already set when the file names it, and, when a DVD file holds several, for the Track. Click Read as text. A progress bar follows the reading, and closing the window stops it.
The subtitles open as a new project saved beside the file, as an SRT of the same name. The pictures are never written over, nor an SRT already there: that one gets a number, like film (2).srt.
A track in a video. Open the video, then in the Import tab choose Subtitles in the video. Picture tracks show as pictures, read with OCR in the Subtitle track list. Pick one and click Import.
A subtitle repeated in back-to-back pictures, as DVDs often do, comes in as one subtitle. Check the text before you save: OCR mixes up some letters that look alike, and italics or unusual fonts are read less well.
Models
There is one model per script, chosen from the language of the subtitles. They are the mobile recognition models of PP-OCRv5, in the ONNX build RapidAI publishes, inside the add-on and checked against pinned fingerprints when it is built: nothing downloads when you use it.
| Model | Size | Good for |
|---|---|---|
| English (en_PP-OCRv5_mobile_rec) en | English | |
| Latin script (latin_PP-OCRv5_mobile_rec) Default latin | French, German, Spanish, Portuguese, Italian, Polish, Turkish and 30 more | |
| Chinese and Japanese (ch_PP-OCRv5_mobile_rec) ch | Chinese, Simplified and Traditional, and Japanese | |
| Korean (korean_PP-OCRv5_mobile_rec) korean | Korean | |
| Russian, Ukrainian, Belarusian (eslav_PP-OCRv5_mobile_rec) eslav | Russian, Ukrainian and Belarusian | |
| Other Cyrillic script (cyrillic_PP-OCRv5_mobile_rec) cyrillic | Bulgarian, Serbian, Macedonian, Kazakh, Mongolian and more | |
| Greek (el_PP-OCRv5_mobile_rec) el | Greek | |
| Arabic script (arabic_PP-OCRv5_mobile_rec) arabic | Arabic, Persian, Urdu, Pashto, Kurdish and more | |
| Devanagari (devanagari_PP-OCRv5_mobile_rec) devanagari | Hindi, Marathi, Nepali, Sanskrit and more | |
| Thai (th_PP-OCRv5_mobile_rec) th | Thai | |
| Tamil (ta_PP-OCRv5_mobile_rec) ta | Tamil | |
| Telugu (te_PP-OCRv5_mobile_rec) te | Telugu |
The add-on runs them without PaddlePaddle or PyTorch. It cuts each picture into its lines and reads each line. Afterwards it puts right what the models get wrong on subtitles: in English, a lone l or | is the word I; in Chinese and Japanese, ?, ! and , after their characters become full-width.
Options
In Settings › Add-ons › PP-OCRv5 (offline OCR) › Configure.
autoWhich model reads the subtitles. Automatic picks it from the language you choose when you open them. Pick one only to override that, for example the Latin script model for Serbian written in Latin letters.
00 uses half of the computer's logical processors, at most 4. Fewer threads keep the computer more responsive while it reads.
Languages
77 languages, in the Latin, Cyrillic, Greek, Arabic, Devanagari, Thai, Tamil and Telugu scripts, Chinese, Japanese and Korean. A language is listed only when every letter of its alphabet is in its model. Romanian (ș, ț), Vietnamese (ơ, ư), Latvian, Welsh, Maltese, Māori, Azerbaijani and Esperanto are left out for that reason, and Hebrew, Georgian, Armenian, Bengali, Khmer, Lao and Burmese have no model. Serbian is read in Cyrillic: for Serbian in Latin letters, choose the Latin script model under Script.
enfrdeesptitnlcagleuplcsskslhrbshutrfisvdanonbnnisetltsqgaidmsswtlfilaflaocuzjvsuzuzhcmnyuejakoruukbebgsrmkkkkymntttgbacvelarfaurpskuugsdhimrnesamaibhonewthtate
Platforms
| System | Download | Tested |
|---|---|---|
| Linux x86-64 | 149.7 MB | Read VobSub and PGS subtitles in Subtitld, on a real machine and in CI |
| macOS (Apple silicon) | 132.3 MB | Read the test pictures with every model in CI (Apple Silicon) |
| Windows x86-64 | 127.0 MB | Read the test pictures with every model in CI |
Linux x86-64, macOS on Apple Silicon and Windows x86-64. Each release is built on all three, and on each one CI installs the add-on as Subtitld does and reads its test pictures with every model.
Troubleshooting
unsupported_languageNo model reads the language you chose. If the subtitles are in a language written with Latin letters, choose the Latin script model under Script.model_missingA model file is missing from the add-on. Reinstall it from Settings ▸ Add-ons.bad_paramsSubtitld couldn't hand a picture over, or an option has a value the add-on doesn't know. The message says which.oomThe computer ran out of memory. Close other programs and try again.internalSomething failed inside the add-on; the message has the details.
If No text was found in these pictures. shows, the track holds no text the add-on can see: some DVD tracks only carry forced subtitles, or are empty.
License
Fine for commercial work: the add-on, ONNX Runtime and the PP-OCRv5 models are all under permissive licences. The text you read from your subtitles is yours.
Versions
- 26.11.1
Offline OCR of picture subtitles for Subtitld: DVD's VobSub and Blu-ray's PGS, as files or as tracks inside a video, read as text with PaddleOCR's PP-OCRv5 mobile recognition models on ONNX Runtime. On the CPU, no torch, nothing downloads.
The archive holds the add-on, ONNX Runtime and all twelve recognition models (Latin script, English, Chinese and Japanese, Korean, East Slavic, Cyrillic, Greek, Arabic script, Devanagari, Thai, Tamil, Telugu), unchanged from RapidAI's ONNX conversion (ModelScope RapidAI/RapidOCR, tag v3.10.0) and checked by sha256.
The add-on and the models are Apache-2.0. Each archive carries
THIRD_PARTY_LICENSES/with the licence of everything frozen into it. The Linux archive carries libquadmath (LGPL-2.1, with numpy); its complete source isppocr-<version>-libquadmath-source.tar.gzin this release.