Fix common subtitle errors
Batch-repair subtitle files in your browser: OCR letter swaps, spacing, unclosed tags, overlaps, impossible durations. Nothing is uploaded.
Subtitles that came out of an OCR pass, a machine translation or a decade-old rip are rarely broken in one interesting way — they are broken in twenty boring ways at once, a few hundred times each. A capital I read as a pipe character, a zero standing in for the letter O, a missing space after every comma, an italic tag that was opened and never closed, two cues that overlap by a third of a second, a line that flashes for 200 milliseconds. Subtitle Doctor runs each repair as a separate, individually switchable rule and tells you exactly how many cues each one touched, so a batch fix is something you can audit rather than something you have to trust. Rules that cannot be applied without guessing at meaning — reading speed, line length, line count — are reported as warnings and never silently rewritten.
Worth knowing
- Every rule is attributed separately in the report: "12 spacing, 4 OCR swaps, 2 unclosed tags" rather than one lumped "file cleaned" count, so you can switch off a rule that is doing the wrong thing to your file.
- The OCR rules refuse to touch real numbers. A zero only becomes an O when it is next to a letter and not next to another digit, which is why "1080p" and "Room 101" survive untouched while "g0 home" is repaired.
- A standalone lowercase L is rewritten as a capital i — but never when an apostrophe follows it, because otherwise every French elision (l’homme, l’amour) would be destroyed by the very pass meant to repair the file.
- Capitalisation after a full stop deliberately skips ellipses: "I waited... and then left" is a continuation, not a new sentence, and uppercasing it is a real error that many batch tools make.
- Timing repairs run in a fixed order — durations first, separation second — so clamping an over-long cue can never reintroduce an overlap that the previous step just cleared.
- Reading speed, line length and line count are measured on visible text only: markup like <i> tags and ASS override blocks is excluded before counting characters.
FAQ
Which errors are fixed automatically and which are only flagged?
Anything with one unambiguous correct answer is fixed: spacing, quotes, OCR letter swaps, dialogue dashes, unclosed tags, repeated lines, empty cues, overlaps and impossible durations. Anything that would require rewriting dialogue — a line that reads too fast, a line that is too long, a cue with too many lines — is listed as a warning with its timestamp so you can decide.
Will this ruin a file that is already in good shape?
No. Every rule is a no-op when there is nothing to fix, and a clean file comes out byte-identical with an empty report. That behaviour is covered by a test in the codebase, precisely because a repair tool that damages good input is worse than no repair tool at all.
Can I turn individual repairs off?
Yes — each rule has its own checkbox, and the timing thresholds (minimum duration, maximum duration, minimum gap) are editable numbers rather than fixed constants. Turning a rule off re-runs the whole pass instantly because everything happens locally in your browser.
What happens to ASS and SSA files?
They come out as SRT. The repairs edit cue text, and edited text cannot be written back into an ASS file without discarding the styling that makes it an ASS file. If you only need timing changes on an .ass file, use the shift or sync tools instead — those preserve the original file byte for byte.