A Subtitle Translation and Review Workflow for Teams
Set up a repeatable subtitle workflow for teams: roles, style guide, glossary, review stages, version naming, automated checks and how to reconcile reviewer changes.
Translating a subtitle file looks simple: the text is already separated from the video, so you just translate it. In practice, SRT files are full of small traps. A translation tool that rewrites the timecodes, merges two lines into one, translates a formatting tag, or produces a sentence twice as long as the original can leave you with a file that will not load, or one that loads and cannot be read in the time available.
This guide gives a reliable workflow for translating an SRT file, whether you do it by hand, with machine translation, or with a mixture of both. It explains what to protect, how to handle longer languages and context, how to check the result, and how to combine a machine draft with a human edit into the best final file.
Subtitle files are creative work. If you wrote the subtitles, or you have permission, go ahead. If you are translating someone else's file, check any licence or terms attached to it, and credit the original author when the terms ask for it. For private study, translating a file you own for yourself is generally fine. For publishing, get permission.
An SRT block looks like this:
12
00:01:15,200 --> 00:01:18,000
<i>I told you not to
open that door.</i>
Only one part of that block is for translators: the text. Everything else is structure that software depends on.
01:15,200 can break the file.<i> and </i> control italics. Keep them around the same words, and keep them paired.Our SRT file guide explains each part in detail. If you remember only one rule, it is this: translate the words and leave the structure alone.
Start from a good original. Fix obvious typos, make sure the file is saved as UTF-8, and confirm it is in sync with the video. Translating errors faithfully only multiplies them. Our guides on encoding and sync help with the technical checks.
A subtitle editor shows the video, the original text, and an empty field for your translation in each row, so you cannot damage the structure. This gives the best quality, and it is the standard way professionals work. Free tools such as Subtitle Edit and Aegisub support this kind of workflow.
You can extract the text lines, translate them, and put them back. This is risky if done manually, because it is easy to lose alignment. If you take this route, keep one subtitle per row, so every original line has exactly one translation, and check that the count matches at the end.
Automatic translation is fast and, for common language pairs, often good enough for a first draft. Subtitles are a hard case for it, though, because each cue is a short fragment taken out of context. Treat the machine's output as a draft that a human must review. Note that if you send text to an online service, the text leaves your device, so do not use it for confidential material unless you are comfortable with the provider's terms.
If you use a tool that accepts a whole SRT file, check afterwards that the timecodes and numbering are untouched. If you paste only the text, use a separator that the translator will not change, such as a numbered list, and translate in chunks of a manageable size.
A subtitle often makes no sense alone: "It's in there" depends on what "it" is. Translators, human or machine, need context.
Languages differ in length. Many translations are longer than the English original, and some are much shorter. A translation that is accurate but too long to read in the time available fails the viewer.
Use these checks:
Languages with different scripts, or text that runs right to left, need extra testing in the player, since line breaking and punctuation can behave differently.
Keep italics for the same things as in the source: off-screen voices, thoughts, foreign words. Apply the target language's punctuation rules: Spanish opens questions and exclamations with inverted marks, French places a space before certain punctuation, and quotation marks and number formats vary. Keep the style of dashes for dialogue consistent.
Do a systematic review, not just a read-through.
A very efficient workflow is to let a machine produce a first draft, then have a person correct it. You then have two versions of the same file: the raw draft and the edited one. To keep every improvement without losing lines that were already fine, compare them in the SRT Merger. Put the edited version in File 1 and the raw draft in File 2, then review the rows where they differ. Identical rows need no attention, which usually shrinks the review a lot.
The same tool is useful when two translators work on different halves of a film or when a reviewer returns a revised file. Our guide to comparing subtitle translations gives criteria for picking between versions.
Translating the timecodes or tags. This corrupts the file. Protect structure.
Letting numbering drift. If lines are merged or split during translation, the cue count changes, and aligning back to the original becomes painful.
Translating every cue in isolation. The result reads like a dictionary, not a conversation.
Ignoring length. An accurate but unreadable subtitle is a failure.
Skipping the playback test. Many problems show up only in motion.
Forgetting names and terms. Inconsistent spellings of a character's name look sloppy. Use a glossary and search the final file.
Losing the encoding. Save the final file as UTF-8 so accents and scripts display correctly.
Translating an SRT file is about translating only the text while protecting everything that makes it a working subtitle file. Start with a clean source, choose a method that matches your quality needs, give the translator context, fit each line into the time available, and check the structure and playback before you finish. If you use machine translation, treat it as a draft and merge it with human edits using a side-by-side comparison. A careful process gives you subtitles that load everywhere and read naturally.
Merge your own subtitle files
Compare two .srt files side by side and save the best lines as one file. It runs in your browser, so nothing is uploaded.