A bilingual subtitle file is one file with two languages stacked in each cue: the original on top, the translation underneath. Players that support it let a viewer read both at once, which is what language learners want and what some broadcasters require for accessibility.
Producing one sounds like a formality — take the timestamps from one file, the text from both, done. It goes wrong often enough that the failure has a recognisable shape: the two languages stop corresponding partway through the film, and by the end every line is matched to the wrong one.
Two ways to match the cues
There are exactly two ways to decide which translated line goes with which original line, and the choice matters more than any other decision in the process.
Matching by position assumes cue 7 in one file is cue 7 in the other. It is exact and it is fast, and it is correct only when both files have the same number of cues in the same order. When that holds — two files derived from the same source — it is perfect. When it does not, it produces a result that looks plausible and is completely wrong.
Matching by time compares each cue's start time and pairs the closest ones within a tolerance. It is slower and it can be fooled by cues that start at nearly the same moment, but it survives the case that breaks positional matching: two files whose cue counts differ.
Why the cue counts differ so often
Two subtitle files for the same video, in different languages, almost never have the same number of cues. This surprises people, because the dialogue is the same dialogue. The reasons are all ordinary:
- The translation splits one long line into two shorter ones, because the target language needs more characters. This is the most common cause by a wide margin.
- The translation merges two short lines into one, because the target language expresses the same thing more compactly.
- One file includes on-screen text like signs and chapter titles, the other omits it.
- One file uses italics for off-screen dialogue and splits those cues differently.
- The two files were timed by different people against slightly different edits of the video.
Once a single split or merge happens, every cue after it is off by one position, and the error compounds: one extra cue early in the film means every subsequent line is matched to the one before it. By the end, the two languages have nothing to do with each other.
What the failure looks like
Suppose the English file has three cues and the Chinese file has four, because the translator split the first line:
English Chinese
1 00:01 Hello there 1 00:01 你好
2 00:04 How are you 2 00:02 那里
3 00:07 Fine thanks 3 00:04 你好吗
4 00:07 很好谢谢
Merged by position, the result is:
1 00:01 Hello there / 你好
2 00:04 How are you / 那里 ← wrong
3 00:07 Fine thanks / 你好吗 ← wrong
4 00:07 很好谢谢 ← orphan
Every line after the split is wrong, and the last translation has no original at all. The file plays without error. Nothing about the output looks broken unless you can read both languages.
Matching by time
Matching by time fixes the case above, because it pairs cues whose start times agree rather than cues whose positions agree. The tolerance — how far apart two start times may be and still count as the same cue — is the setting that matters.
| Tolerance | Behaviour |
|---|---|
| Under 100 ms | Only cues timed identically match. Safe, but misses cues that were retimed slightly. |
| 100-250 ms | The usual working range. Matches cues that were nudged by a frame or two. |
| Over 500 ms | Starts pairing cues that are genuinely different lines, especially in fast dialogue. |
A tolerance around 100 to 250 milliseconds is right for most files. Larger values do not help with the split-cue problem — a split cue has a genuinely different start time, not a slightly different one — and they introduce a new failure where two unrelated short lines get paired because they happen to be close together.
The honest limitation: time-based matching cannot attach a split half to the cue it belongs with. It will match the first half to the original line, and the second half has no counterpart, so it becomes a cue with only one language. That is a visible, comprehensible outcome — far better than a silently mismatched file.
What to do with cues that have no counterpart
Any merge between files of unequal length produces cues with text in one language and nothing in the other. There are three ways to handle them and each is right in a different situation:
- Keep them, with one language only. The viewer sees a line in one language for that moment. Honest, and usually the best choice: dropping the text loses information the original had.
- Merge them into the neighbouring cue. Combines two lines into one cue, which may overflow the reading limits — worth checking the CPS afterwards.
- Drop them. Only defensible when the orphan is on-screen text that genuinely has no translation, not when it is dialogue.
Whatever you choose, the output has to stay in time order. A cue that was orphaned in the middle of the film belongs in the middle of the film, not appended after the last cue — an out-of-order file is rejected by some players and shows as corrupt in most editors.
The readability problem nobody mentions
A bilingual cue carries twice the text in the same time slot. A line that was comfortable at 15 characters per second in one language is now 30, which is past the point where anyone can read it — and past the point where the two lines fit on screen.
This is the structural problem with bilingual subtitles and no merge tool can solve it, because the fix is in the source files:
- Reduce each language to one line rather than two. A bilingual cue with four lines total occupies most of the picture.
- Keep each line short. Two languages at 42 characters each is 84 characters in the cue; that is already the limit for a two-line cue.
- Expect to condense the translation. A bilingual subtitle is not a faithful subtitle — it is a reading aid, and it has to be shorter than a monolingual one to work.
Try it on your own file — no upload, nothing leaves your browser.
Open the Merge Two Subtitles into Bilingual SRT/VTT - Free Tool →After merging, the file is worth checking for readability like any other — the merge does not create timing problems, but it does double the text that has to be read in the same window.
A procedure that holds up
- Check the cue count in both files before merging. If they match, positional matching is safe and exact. If they differ, use time-based matching.
- Use a tolerance of 100-250 ms. Do not raise it to force more matches — a forced match is a wrong match.
- Merge, then look at the cues that came out with one language only. Those are the places the two files disagree, and they are where a human decision is needed.
- Confirm the output is in time order and every cue has a sequential number.
- Check readability afterwards. The merged file has roughly twice the characters per cue as either source.
The first step is the one people skip, and it is the one that decides whether the other four matter.