Should you merge overlapping subtitle cues?

Merging can preserve words while changing their timing

Do not automatically merge overlapping subtitle cues into one block spanning the earliest start and latest end. That keeps all the text but can reveal one speaker too early and leave another visible too long. First establish whether the overlap belongs in the recording; then choose a delivery representation that preserves the intended meaning and remains readable.

This guide starts after you have confirmed that two or more cues in one subtitle track share time. If you are still diagnosing why captions appear twice, use the double-caption guide first. Here the question is narrower: what happens to the words and display intervals when you keep, trim, delay, merge or partition those cues? Work only with media and files you own or are authorized to edit.

Keep a reference for the timing decision

Save an untouched subtitle export and identify the exact video revision, language and working copy. Preserve any unsaved edits before switching files. Note the overlapping cues’ original identifiers, complete text and timing. Review from a few seconds before the passage until after both cues clear, with temporary player subtitle delays disabled.

Listen to the recording and check any recent structural edit before choosing an operation. Aegisub, for example, documents a preserve-times split that gives both resulting lines the old interval. That can create an accidental overlap. If the original voices are sequential, find their actual phrase transition rather than applying a generic merge.

W3C’s WebVTT specification permits overlapping cues; chapter cues have separate restrictions. A valid temporal overlap is therefore not automatically an error, although an SRT destination may present it differently. W3C’s caption guidance also requires attention to meaningful non-speech information. Keep every necessary voice and sound in the review.

Measure the shared interval

For two cues, take the later start and the earlier end. If the earlier end is after the later start, the difference is the overlap duration. Equal endpoints mean the cues touch without sharing a positive duration; a gap means they do not overlap. This calculation identifies a review region, not the correct speech boundaries.

Consider this invented scene. Mira’s caption runs from twenty to twenty-four seconds. Leon’s begins at twenty-three seconds and ends at twenty-five. Assume, for this example, that listening confirms Leon interrupts while Mira is still speaking. The shared display interval is twenty-three to twenty-four seconds: one second. The labels below are manually supplied teaching examples, not automatic speaker identification.

31
00:00:20,000 --> 00:00:24,000
MIRA: Leave the door open.

32
00:00:23,000 --> 00:00:25,000
LEON: Wait!

Why the obvious repairs can change the scene

Trimming cue 31 to end at 23 seconds removes one second of its display. That may be justified if the original end was wrong, but the overlap calculation alone cannot establish it. Delaying cue 32 to 24 seconds avoids the overlap by presenting Leon’s interruption one second later. It changes the relationship between the captions and the assumed audio.

Combining both texts into one cue from 20 to 25 seconds creates a five-second shared display. Leon’s line then appears three seconds before its original start, and Mira’s remains one second beyond its original end. A merged cue can contain every word and still tell the scene at the wrong moments. Do not treat a larger combined interval as a neutral cleanup.

Caption display times are not necessarily exact speech onset and offset measurements. Reading time and the receiving specification also matter. Use the audio and picture to judge a proposed change, then check readability at normal speed. Do not impose a universal gap, minimum duration or frame rule on every SRT file.

Choose a response the evidence supports

Keep an intentional overlap when it represents the scene accurately and the intended player presents it readably. Check that speaker labels remain understandable and that important visuals stay visible. A format accepting the timestamps is only the first check; the viewer must be able to follow the result.

Correct an accidental overlap by changing the boundary that is actually wrong. For a lingering first caption, review its end. For an early second caption, review its start. For an unintended preserve-times split, find a real phrase transition and retime both pieces deliberately. Preserve the unaffected text and neighboring cues, and record why the adjustment was made.

If the destination cannot present the intentional overlap clearly, prepare a separate delivery adaptation. Its requirements may call for a reviewed two-speaker caption or a different supported format and layout. That is an editorial and timing decision, not a lossless conversion. A reviewer needs to check word order, reveal time and reading load. Keep the original reference so the adapted delivery is not mistaken for the untouched track.

A sequential alternative needs more than one combined block

If a destination needs one displayed block at a time, you can consider partitioning the original intervals at every start and end boundary. In our example, the display states are Mira alone from 20 to 23 seconds, both speakers from 23 to 24 seconds, and Leon alone from 24 to 25 seconds. The following is an illustrative candidate, not a universal recommended delivery:

Every speaker’s original text remains present for the same total interval, but the middle block lasts just one second and each line continues across a block transition. A player may reposition or redraw the words, and viewers may interpret repeated lines as repeated speech. The candidate therefore preserves the interval model without proving readable playback. Reject or revise it if those transitions distract or the destination specification disallows them.

This partition changes two original cues into three delivery blocks. Record old 31 as contributing to new 31–32 and old 32 to new 32–33. Renumber subsequent blocks without shifting their timing. Keep this mapping outside the subtitle text, and retain the original overlapping version. A single source cue no longer corresponds to a single delivery cue, which matters when comparing or updating translations.

31
00:00:20,000 --> 00:00:23,000
MIRA: Leave the door open.

32
00:00:23,000 --> 00:00:24,000
MIRA: Leave the door open.
LEON: Wait!

33
00:00:24,000 --> 00:00:25,000
LEON: Wait!

Check the whole overlapping group

Do not check only the next cue and assume the rest are clear. A long cue from 10 to 20 seconds overlaps a short cue from 12 to 13 and another from 18 to 19, even though those two short cues do not overlap each other. Inspect every cue active within the passage, including long sound descriptions. A fix to one pair can leave another collision unresolved.

Keep a small external review note with the original cue IDs, cause, approved timing change or keep decision, and destination result. If you combine or split blocks externally, retain an old-to-new identifier map. Cue counts should change only for an intentional structural edit; a correct count alone does not prove that every word or timestamp survived.

Review each translated track separately. Longer text or different word order can make an acceptable source-language overlap difficult to read. Preserve completed translations and manual edits, and have someone who understands the target language review the affected passage. Do not assume a timing change in one language safely fixes every other track.

Reopen the export and test the delivered track

Polyglot preserves overlaps when importing an SRT and supports reviewing text, editing timing and exporting a separate subtitle file. Use those capabilities for a deliberate review; this workflow does not rely on automatic overlap repair, cue merging or speaker recognition. Structural changes can be prepared in a suitable external editor while the original project and exports remain safe.

Reopen the export and compare the affected text, identifiers and timestamp pairs with your review note. The Library of Congress describes the SRT block structure: counter, start/end timing, subtitle text and a blank separator. Confirm that each cue still has a positive duration and that unrelated passages have not changed.

Finally, select the exact exported language track with the final video in the intended player. Watch before, during and after every reviewed overlap at normal speed, including a smaller viewing size. Confirm that neither speaker disappears, no line flashes too briefly, text stays readable and important picture content remains visible. Then review the complete track. If only a local player was checked, keep destination acceptance pending until the actual delivery is reviewed.

Keep exploring.