Subtitle Edit

the subtitle editor :)

OCR (Optical Character Recognition)

Subtitle Edit can convert image-based subtitle formats to text using OCR.

OCR Window

Supported Image Formats

OCR Engines

Tesseract

Open-source OCR engine with language packs.

nOCR (Nikse OCR)

Built-in trainable OCR engine.

Binary OCR

Binary image comparison engine, listed as Binary image compare in the engine dropdown.

Apple Vision

Built-in OCR on macOS using Apple’s Vision framework.

Google Lens Sharp / Google Lens Standalone

Cloud-based OCR using Google Lens (free, but capped).

Google Vision

Cloud-based OCR using Google Cloud Vision API.

Ollama

Local LLM-based OCR via an Ollama server (e.g. with a vision model).

llama.cpp

Local LLM-based OCR using a llama.cpp-compatible server. Subtitle Edit can download llama.cpp and the model for you, or talk to a server you run yourself.

CrispEmbed

Local OCR engine with multiple model backends (free/open source).

Mistral OCR

Cloud-based OCR using Mistral API.

Paddle OCR Standalone / Paddle OCR Python

Local OCR engine, based on Paddle OCR 3.7 with the PP-OCRv6 recognition models.

Engine Setup Notes

How to Use

  1. Open an image-based subtitle file
  2. The OCR window opens automatically
  3. Select an OCR engine
  4. Configure engine-specific settings
  5. Click Start OCR
  6. Review and correct any errors
  7. Click OK to import the text subtitles

After OK, the matching video is opened as well (unless auto-opening the video is turned off in the settings): a container source (.mkv, .mp4, .ts) is opened as the video itself; for a stand-alone image subtitle (.sup, .sub/.idx, …) a video passed together with the file (drag and drop, command line) is used first, then the video the subtitle was last opened with, then a video file with the same name next to the subtitle.

Language Auto-detection

When the OCR window opens, the OCR language for every engine - and through it the spell check dictionary - is pre-selected from the source’s declared language: the stream language in a VobSub .idx, the track language of a Matroska or MP4 subtitle track, or a language tag in the file name. Two file-name patterns are recognised: a bracketed ISO 639 code such as movie_track3_[dut].sup (the naming used by Subtitle Edit’s own Matroska track export; the last valid code wins), and a trailing dot-separated tag such as movie.nl.sub or movie.dut.forced.sub - the hi, sdh, cc and forced markers are skipped, only the last two remaining tokens are considered, three-letter codes must be lowercase, and hi is read as hearing impaired rather than Hindi. A detected language wins over the last-used one; a source without language information keeps the previous selection.

Options

Checkboxes below the engine settings:

Toggle buttons above the image:

Subtitle List Menu

Right-click the subtitle list for:

Find and Replace

Find… and Replace… (subtitle list or text box right-click menu, Ctrl+F / Ctrl+H) open the same Find and Replace windows as the main window, so text can be searched and fixed while the image is still next to it. A match selects its line and highlights the text in the text box. Replace all also updates the lines’ highlighted OCR text in the list.

Saving the Images

The subtitle list’s right-click menu has Save all images with HTML index…. Pick a folder and Subtitle Edit writes index.html plus images/0001.png, 0002.png, … — every subtitle bitmap next to the text OCR produced for it, which is the quickest way to proof-read a run against the originals.

The page is self-contained (no external css, js or fonts), so it opens straight off the file system. It follows the reader’s light/dark preference, with an Auto/Light/Dark override, and adds a text filter, an “only lines without text” filter for spotting images that produced nothing, a dark/light/checkerboard backdrop for the transparent bitmaps, click-to-zoom, and each line’s number, time codes, duration and image size.

Keyboard Shortcuts

General

Shortcut Action
Escape Cancel OCR / Close window
Ctrl+G Go to line number
Ctrl+F Find
Ctrl+H Replace (Cmd+Alt+F on macOS)
F3 / Shift+F3 Find next / Find previous
Ctrl++ Zoom in (images in grid)
Ctrl+- Zoom out (images in grid)
F1 Show help (default key, changeable)

Subtitle Grid

Shortcut Action
Ctrl+I Toggle italic formatting
Ctrl+P View selected image (use arrow keys to navigate)
Delete Delete selected line(s)
Home Jump to first line
End Jump to last line
Double-click Inspect line (nOCR/Binary OCR only)

Unknown Words List

Shortcut Action
Enter Jump to subtitle line containing the selected unknown word

Note: All shortcuts can be customized. Go to Options → Shortcuts to view and change key bindings.

Pre-processing

Before OCR, you can apply image pre-processing via the Image pre-processing toggle button in the OCR window (it shows as pressed while any pre-processing is active):

The settings are remembered between sessions and applied automatically the next time you open the OCR window. Reset in the Pre-processing window sets all options back to their defaults (all off, border size 2, darkness threshold 128).

Batch OCR

Batch Convert can process image-based subtitle files with OCR. Subtitle Edit 5 includes Binary OCR in Batch Convert and can auto-detect several nOCR/Binary OCR settings such as language and pixels-are-space values.

For VobSub files, Batch Convert isolates the glyph color before OCR (on by default, configurable in the Batch Convert settings): the subpicture is rebuilt as a crisp black-on-white bitmap from pixel frequency, so outline and anti-alias colors no longer melt adjacent characters together.

For command-line workflows, see Command Line (seconv), which documents OCR engines and options for headless conversion.

OCR Pre-processing

Unknown Words

When the OCR engine encounters uncertain characters, you can:

The unknown-words list in the OCR window has buttons for the selected word: Add to names list (case sensitive), Add to user dictionary, Add to OCR replace pairs and Google it.

What is remembered

Some of the choices in the unknown-word prompt are saved to disk and reused on later OCR runs; others only last as long as the OCR window is open.

Choice Remembered? Stored in
Change all Yes {language}_OCRFixReplaceList.xml
Add to names list Yes {language}_names.xml
Add to user dictionary Yes {language}_user.xml
Skip once No —
Skip all No — current OCR session only —

“Skip all” is deliberately temporary. It silences a word for the rest of the current OCR session and is forgotten when the OCR window closes, so a mis-click never has lasting consequences. If you want a word to be accepted permanently, use Add to user dictionary (or Add to names list for proper nouns) instead, and use Change all for a correction that should be applied automatically from now on.

OCR Fix Replacement Lists

Subtitle Edit uses language-specific XML files to automatically correct common OCR errors. These files are named {language}_OCRFixReplaceList.xml (e.g., eng_OCRFixReplaceList.xml for English) and are located in the Dictionaries folder.

See the example file: Dictionaries/eng_OCRFixReplaceList.xml

XML Structure

The OCR fix replacement list contains several sections that handle different types of corrections:

1. WholeWords

Replaces entire words that match exactly. This is the most common section for fixing OCR mistakes.

<WholeWords>
    <Word from="tñere" to="there" />
    <Word from="ri9ht" to="right" />
    <Word from="0f" to="of" />
    <Word from="alot" to="a lot" />
    <Word from="becuase" to="because" />
</WholeWords>

Use cases:

2. PartialWordsAlways

Replaces character sequences within words, always applied without spell checking.

<PartialWordsAlways>
    <WordPart from="¤" to="o" />
    <WordPart from="lVI" to="M" />
    <WordPart from="IVl" to="M" />
</PartialWordsAlways>

Use cases:

3. WholeLines

Replaces entire lines that match exactly (including formatting tags).

<WholeLines>
    <Line from="[chitte rs]" to="[chitters]" />
    <Line from="Hil' it!" to="Hit it!" />
    <Line from="&lt;i&gt;Hil' it!&lt;/i&gt;" to="&lt;i&gt;Hit it!&lt;/i&gt;" />
    <Line from="ISIGHS]" to="[SIGHS]" />
</WholeLines>

Use cases:

4. PartialLinesAlways

Replaces text fragments within lines, always applied without spell checking.

<PartialLinesAlways>
    <LinePart from="Apollo 1 3" to="Apollo 13" />
    <LinePart from=",.," to="..." />
    <LinePart from=" lt " to=" it " />
    <LinePart from=" lf " to=" if " />
</PartialLinesAlways>

Use cases:

5. PartialLines

Replaces text fragments within lines (may be spell-checked).

<PartialLines>
    <LinePart from=" /be " to=" I be " />
    <LinePart from=" aren '1'" to=" aren't" />
    <LinePart from=" aren'tyou" to=" aren't you" />
</PartialLines>

6. RegularExpressionsIfSpelledCorrectly

Uses regex patterns to fix errors, but only applies the replacement if the corrected word is in the dictionary.

<RegularExpressionsIfSpelledCorrectly>
    <!-- Fix lowercase 'l' to uppercase 'I' if result is a valid word -->
    <RegEx find="\bl([A-Z]+)\b" spellCheck="I$1" replaceWith="I$1" />
    <RegEx find="\b([A-Z]+)l\b" spellCheck="$1I" replaceWith="$1I" />

    <!-- Fix possessive forms: David's, there's -->
    <RegEx find="\b([A-Z][a-z]+)['']s\b" spellCheck="$1" replaceWith="$1's" />
    <RegEx find="\b([a-z]+)['']s\b" spellCheck="$1" replaceWith="$1's" />

    <!-- Fix missing spaces: ofDavid → of David -->
    <RegEx find="\bof([A-Z][a-z]+)\b" spellCheck="$1" replaceWith="of $1" />
    <RegEx find="\bin([A-Z][a-z]+)\b" spellCheck="$1" replaceWith="in $1" />

    <!-- Fix 'l' in brackets: [GRlNDlNG] → [GRINDING] -->
    <RegEx find="\[([A-Z ]*)l([A-Z ]*)\]" spellCheck="[$1I$2]" replaceWith="[$1I$2]" />
</RegularExpressionsIfSpelledCorrectly>

Attributes:

Use cases:

Creating Custom Rules

To add your own OCR fix rules:

  1. Open the appropriate language file: Dictionaries/{language}_OCRFixReplaceList.xml
  2. Add entries to the relevant section based on the type of error
  3. Save the file
  4. Restart Subtitle Edit for changes to take effect

Tips: