Subtitles showing twice? Find the extra layer first

Isolate what is drawing the second caption

When subtitles appear twice, first turn off the player’s selectable captions and replay the same moment. Note what disappears. Then identify whether the remaining words are part of the video picture, another caption overlay or content inside the selected subtitle file. Do not start by deleting every repeated sentence or removing all overlapping cues.

Use this workflow with media and subtitle files you own or are authorized to inspect and change. Save untouched copies before editing. The tests below are diagnostic steps, not a claim that every player behaves identically or that Polyglot can automatically repair double captions.

Record one reproducible moment

Choose a short interval where the problem is clear. Write down the video filename or page, video revision, selected language, subtitle filename if known and playback time. Describe whether the two copies have identical words, different languages or different timing. A screenshot helps document appearance, but the original audio and timed text are needed to judge the cause.

Replay a few seconds before and after that moment at normal speed. Does the duplication persist for the whole video, or only while two captions change over? A persistent second copy suggests a display-layer problem; a brief collision suggests a cue boundary worth inspecting. These are clues, not conclusions. Two intentionally displayed languages may also look like duplication to someone expecting one.

Turn off one display source at a time

Use the player’s caption control first. YouTube’s official help, for example, documents clicking its Closed captions control to turn captions off. That changes the player’s selectable caption display; it does not edit an uploaded SRT or erase text from the video picture. For another player, follow its own current controls.

If one copy disappears and another remains, you have isolated at least one display source. Do not immediately conclude that the remaining copy is burned in. Check whether a browser extension, separate live-caption window or another explicitly enabled overlay is displaying it. Temporarily hide one known source at a time and repeat the same interval, keeping a note so you can restore the original settings.

If both copies disappear with the player captions, inspect the selected track’s text and timing next. That result is consistent with repeated text or simultaneous cues in the track, but it is not proof: the player may manage more than one display element together. If neither disappears, confirm that you used the control for the player actually showing the video.

Distinguish video pixels from a selectable track

W3C distinguishes closed captions, which viewers can hide, from open captions, which remain displayed. When captions have been rendered into the video pixels, disabling a subtitle track cannot remove those pixels. Compare the exact delivered video with a known clean master in a player with selectable captions and known external overlays off.

If the delivered picture already contains the same caption that a sidecar file adds, choose the intended deliverable with the owner. A clean video plus a reviewed selectable track preserves viewer control. An intentionally captioned video may be required by a particular destination. Do not delete the only complete accessibility track merely to make a screenshot look tidier: the embedded text might omit meaningful sounds, contain another language or cover only selected lines.

To remove captions that are part of the picture, return to the clean source and the video-editing project when available. Editing or exporting an SRT cannot recover the image underneath them. Polyglot does not promise to remove burned-in text or render a replacement burned-in video.

Inspect the file at the exact interval

Open a copy of the selected SRT or WebVTT as plain text or in a suitable subtitle editor. Find the affected words. Check whether they occur twice inside one cue, in two cue blocks with intersecting times, or at separate moments where the speaker actually repeats them. A search result containing the same sentence twice is not sufficient evidence for deletion.

For two cues, the shared interval starts at the later start time and ends at the earlier end time. If that end is later than that start, the cues overlap. In this fictional SRT excerpt, both cues are active from 12.500 to 13.000 seconds: a half-second overlap. The words differ, so deleting either would lose content.

Listen to the matching recording. If the first phrase ended before the second began and its caption simply lingers, adjust the confirmed boundary in a working copy. If the voices really overlap, preserve that meaning and review how the destination presents both. The example alone supplies no evidence for a correct replacement timestamp.

17
00:00:10,000 --> 00:00:13,000
Hold here.

18
00:00:12,500 --> 00:00:14,000
Now release.

Keep intentional overlap and repetition

The W3C WebVTT specification explicitly permits different cues to overlap. Valid timing therefore does not guarantee a comfortable visual result, and an overlap warning does not by itself identify an error. Subtitle tracks are also different from WebVTT chapter tracks, which have their own restrictions; do not generalize a chapter rule to all captions.

Simultaneous speech, meaningful sound and an intentional bilingual presentation can require more than one piece of text. Check speaker attribution and whether a supposedly duplicated line belongs to another voice. Preserve a real repeated word, refrain or instruction unless the agreed editorial style calls for a reviewed change. W3C’s caption guidance includes speech and the non-speech audio needed to understand the content.

If two blocks have identical text and timing, compare them with the untouched file, recording and any approved revision before treating one as accidental. Keep a record of every removed block. Renumbering cues alone will not remove repeated text, and shifting the entire track will not resolve the overlap between two cues shifted by the same amount.

Make one repair and verify the complete result

Match the repair to the evidence: disable an unintended overlay, select the intended track, use the agreed video derivative, remove a confirmed accidental duplicate in a suitable editor or correct a verified cue boundary. Change one cause at a time so you can tell which action worked. Keep the original exports and unsaved work safe.

Polyglot can import an SRT, review and edit its text, make explicit timing adjustments and export a separate subtitle file locally. Imported overlaps are preserved rather than automatically treated as mistakes. This guide does not promise a duplicate-removal button, automatic speaker recognition or management of another player’s overlays. After changing text or timing, review affected translations separately.

Load the final file with the exact delivered video in the destination player. Restore the intended caption settings and check the reported interval, both neighboring cues, repeated dialogue elsewhere and any intentional simultaneous content. Then review the complete track. Verify that viewers still have the required words and language, including meaningful sounds where needed; simply making the second copy vanish is not enough.

Keep a short external change note with the original symptom, isolated cause, precise repair and destination result. If you checked only a local player, leave the destination check pending. The note below is a manual review aid, not an automatically generated Polyglot report.

Video revision: ______
Caption file: ________
Time checked: ________
Extra layer: _________
Confirmed cause: _____
Repair made: _________
Words retained: ______
Final player: ________
Pending checks: ______

Keep exploring.