Split a long subtitle cue without guessing the timing

Divide the words and the time together

To split a long subtitle cue, choose a sensible phrase boundary, find that boundary in the matching audio, and create two cue blocks with separate start and end times. Preserve the original words and check the new cues alongside their neighbors. Dividing the text in half does not tell you when the speaker reached the middle.

A cue is one timed block of subtitle text. Adding a line break inside that block still shows all its words during the same interval. Splitting creates two successive displays. Use a line break when the problem is wrapping; consider a cue split when the viewer should receive the information in stages. This guide uses recordings and subtitle files you own or have permission to edit.

Keep a reference before changing the structure

Save the original SRT and identify the exact video revision it matches. Work on a copy. Write down the affected cue’s identifier, text, start and end, plus the neighboring cues. Keep any unsaved editor changes safe before exporting or opening another version.

Check the recording at normal speed. Is the caption holding two complete phrases for too long, revealing a later instruction early, or spanning a change of speaker? A split can help with those problems. If the whole track is uniformly late, splitting a cue is the wrong repair; diagnose synchronization first. If words are wrong, correct them as a separate documented change.

Choose a division that keeps meaning intact. Avoid leaving an article detached from its noun, separating a name, or putting a negation on a different display from the phrase it qualifies. A sentence boundary is a useful candidate, but the audio must still support the timing. A short cue that flashes past can be worse than a longer, readable one.

Find the boundary in the speech

Listen from before the cue until after it. Locate the end of the first phrase and the beginning of the next, using repeated playback and the editor’s audio or video controls. A pause can provide space for a transition, but a pause alone does not tell you whether the previous caption should remain visible. Consider the picture, the words and the receiving platform’s requirements.

Do not assume equal word counts deserve equal time. A speaker may deliver a short first instruction quickly and explain the second slowly. Text-length estimates cannot detect that difference. W3C’s caption guidance treats synchronization with audio and accurate content as essential; mechanically smaller blocks are not enough.

For continuous speech, a shared boundary can give one cue’s end and the next cue’s start the same elapsed time. If your destination specifies a frame gap, duration limit or shot-change rule, follow that specification using the actual video timebase. There is no universal millisecond gap to invent for every SRT. Review any deliberate gap or overlap against the soundtrack.

Know what the split command actually does

Aegisub’s editing documentation gives three distinct split behaviors. Preserving times gives both new lines the original interval. Estimating times divides according to the amount of text on each side. Splitting at the video frame uses the current frame for the second line’s start and the previous frame for the first line’s end. These are documented Aegisub behaviors, not Polyglot controls or a hands-on test here.

The preserved-time option can leave both captions appearing together. The estimate still needs a listening check. The frame-based option depends on where you placed playback. Whatever editor you use, inspect both resulting timestamp pairs immediately and confirm that only the intended cue was changed.

For a very short SRT, you can prepare the two blocks in a plain-text editor after measuring the boundary elsewhere. Keep the counter, timestamp line, text and blank separator for each block. The Library of Congress describes this basic SRT structure. A blank line separates cues; it is not a substitute for the second cue’s number and timing line.

Work through an unequal split

Imagine an authorized lantern-making demonstration. The original cue runs from ten to eighteen seconds. In this invented teaching example, “Fold the paper.” finishes before the second phrase begins at 12.400 seconds. We choose that observed phrase transition as the shared boundary for this example. These timings are supplied assumptions, not measurements from a real recording.

The original block below shows both instructions at once. Its internal line break changes layout but does not postpone the second instruction.

7
00:00:10,000 --> 00:00:18,000
Fold the paper.
Then press along the crease.

Check the two new blocks

The first new cue lasts 2.4 seconds and the second 5.6 seconds. An equal division would place the boundary at fourteen seconds, 1.6 seconds after the assumed start of the second phrase. Even though both methods produce eight seconds in total, they do not reveal the words at the same moments.

This example retains the original ten-second start and eighteen-second end. In a real file, preserve those endpoints only if they are already correct. Review the new durations for comfortable reading in context; the example is not a universal minimum or maximum duration rule.

7
00:00:10,000 --> 00:00:12,400
Fold the paper.

8
00:00:12,400 --> 00:00:18,000
Then press along the crease.

Account for every word and cue number

Compare the two texts in order with the original. All words and punctuation should remain unless you deliberately approved a separate edit. Watch for a lost word at the cursor, duplicated punctuation, a sound description left behind or a speaker label now attached to the wrong phrase. Captions may need meaningful non-speech information as well as dialogue, as W3C explains.

One cue becoming two increases the actual cue count by one. In a file with twenty blocks and no other changes, expect twenty-one afterward. That count is a useful check, not proof of correctness: text can still be lost while the count is right.

If the original next cue was numbered eight, it now needs a new unique number. Renumber the delivery in order and keep a small external mapping: old 7 becomes new 7–8; old 8 becomes new 9. All later cues keep their original text and times unless another change is documented. Changing cue numbers does not require shifting the rest of the soundtrack.

Treat translations as their own review

A translated sentence may use a different word order or need a different phrase boundary. Do not split every language after the same word count. Have someone who understands the translation review which words belong in each new interval, while retaining the intended meaning and relationship to the audio.

Existing translation references can become confusing after renumbering. Keep the untouched project and exported language files, record the structural change, and review the affected passage in each deliverable. Do not assume an external SRT split automatically restructures saved translations. If a bilingual export depends on corresponding cues, check its alignment again before replacing the delivered copy.

Review the prepared file and its final playback

Polyglot can import an SRT, review and edit its text and timing, and export a separate subtitle file. Prepare the structural split in an editor with a documented split operation or in a careful plain-text working copy, then review the resulting file. This procedure does not depend on a native automatic cue-splitting command.

Reopen the export and verify that both new cues have positive durations, unique identifiers and the intended text. Check the previous cue’s end, the new boundary and the following cue’s start. Listen at normal speed with temporary player subtitle delays disabled. Confirm that the first phrase clears when intended and the second is neither prematurely revealed nor delayed.

Finally, load the delivered file with the exact video in the intended player. Check reading comfort, accented characters and language selection, and review the complete track for accidental changes. If only local playback was checked, leave destination acceptance pending. Keep the original, the revised export and the cue mapping until the recipient confirms the correct version is in use.

Keep exploring.