Anjin Media
All field notes

Delivery checks

How to fix video caption errors before publishing

Find and fix caption errors in burned video, subtitle files and platform tracks, with a worked correction across two campaign versions.

BY ANJIN MEDIA EDITORIAL7 MIN READ

Correcting an SRT file does not correct a spelling mistake already burned into a video. Correcting the video does not necessarily replace the caption track uploaded to a platform.

Before fixing a caption error, identify where the text comes from: the picture itself, a separate caption file or a platform-generated track. Check the intended words against the audio and source context. Repair the affected output, then test the exact file or player the audience will use.

For in-house marketers making several edits from one webinar or interview, the useful question is not only “is this word right?” It is “which versions still contain the wrong word?” This workflow turns a small correction into a controlled delivery decision without requiring another approval form.

01

Locate the caption layer

Open the actual video file in a player and check whether captions remain visible when the player's optional track is off. Text that is part of the picture cannot be switched off as a separate track. A second set of words appearing when captions are enabled is a sign to investigate two layers, not proof that one is correct.

Adobe's export settings reference distinguishes burned captions, separate sidecar files and captions embedded as a selectable stream. Use that distinction to ask for the correct repair. Different destinations can display different layers even when they use the same video.

  • Wrong text in the picture: correct the caption source in a workflow that can render the video again, or obtain a clean source and rebuild the caption treatment.
  • Wrong text in a separate track: correct and replace the relevant track, then confirm the player is using it.
  • Wrong platform-generated captions: use that platform's supported correction process, then check whether an uploaded track or burned text also needs attention.

A new caption layer is not a reliable way to hide an old one. It can leave duplicate words, obscure important visuals or expose the old error when the viewer changes settings.

02

Establish the words before editing the file

Listen to the selected passage with its surrounding sentence. Check names and product terminology against an appropriate reference, but do not replace what the speaker said merely because different wording would make a stronger campaign claim.

If the speaker says something incorrect, changing the caption to a correct statement would misrepresent the audio. Treat that as an editorial issue: select a supported passage, include an appropriate correction or hold the claim. The video review guide covers those decisions.

Pay attention to negations, units, decimal points and speaker changes. A caption can look grammatically tidy while reversing the meaning. W3C guidance on captions explains the need to check automatic captions for accuracy and include meaningful audio information.

When the sound is unclear, mark the uncertainty for review. Do not resolve an ambiguous number by choosing the value the campaign team hoped to hear.

03

Work through a correction across two versions

Hypothetical example: a recorded presenter refers to an invented product called “Northline”. Automatic captions turn the name into “north line”. The marketing team makes a square social edit with burned captions and a landscape website edit with a separate subtitle track. This scenario is illustrative, not a customer result.

First, a reviewer listens to the audio and confirms the spelling from the approved product reference. The correction is a transcription fix, not a rewrite of the speaker's claim.

For the square video, the team corrects the caption input and produces a new export. Sending an edited SRT alongside the old square MP4 would leave the visible error unchanged.

For the website version, the team replaces the matching subtitle track and checks it in the actual player. If the video also contains burned captions, that layer needs a new render too. The team does not assume the website and social edits use identical timing simply because they came from the same interview.

The release note can be one sentence in the existing asset tracker: “Product spelling checked against audio and approved reference; square video replaced; landscape caption track replaced and playback checked.” Attach the exact filenames or asset identifiers. The note records what was done rather than a general claim that all future uses of the recording are now correct.

04

Check timing after the wording is right

A correctly spelled caption can still arrive too early, remain over the next speaker or refer to a screen that is no longer visible. Watch the finished edit, especially around cuts. Do not review the text file in isolation.

Two timing patterns need different investigations. If every caption is offset by the same amount, check whether the file belongs to a different export or whether an introduction was added. If alignment gets progressively worse, investigate the timing conversion or mismatched source version rather than shifting a few individual cues until the opening looks right.

When the sequence changes, obtain captions matched to the changed sequence. A subtitle file for the full webinar will not automatically match a short edit assembled from non-adjacent passages.

Use the intended display size to assess reading time, line breaks and placement. Correcting a long product name may change wrapping. A caption that fits a landscape frame may cover a chart in portrait. The slide-heavy webinar guide explains how to keep the visual evidence readable.

05

Keep the corrected files together

Give each delivered video an unambiguous version and pair its caption files with that version. For example, an internal naming convention might use campaign_answer_v4_square.mp4 and its corresponding caption file. The names are illustrative; use the team's existing convention if it already makes the relationship clear.

Keep old files distinguishable from the replacement so nobody accidentally republishes them. Record which destinations have received the corrected asset and which still need updating. A corrected folder on a shared drive does not establish that a scheduled ad, website embed or downloaded sales clip has changed.

If a finishing editor is making the repair, send the verified wording, the reference video, the affected time range and the required outputs. The finishing handoff guide helps separate the reference file from the sources needed to rebuild it.

06

Check Anjin Media outputs individually

Anjin Media documents caption and sidecar outputs. Its results reference describes SRT and VTT captions timed to the finished edit, plus the styled ASS file when there are caption cues to burn. Check the rendered picture and the delivered subtitle files separately; a correct sidecar alone does not establish that every distributed output is correct.

Choose the correction and rendering workflow supported by the tools your team actually uses. Do not assume that editing a downloaded file updates an existing render or a third-party upload. After a repair, inspect the new result and retain the version information needed to distinguish it from the previous export.

A successful technical quality check is not a word-by-word editorial sign-off. The reviewer still needs to check names, meaning, timing and readability in the output intended for the audience.

07

Questions that prevent repeat corrections

Can a glossary replace listening to the recording?

No. A glossary helps with spelling and terminology, but it cannot tell you whether the speaker used that term in this sentence. Use it beside the audio, especially when two product names sound alike.

Should we remove every hesitation from captions?

Use an agreed captioning approach that preserves meaning and follows the delivery requirements. Do not make uncertain speech look certain by removing a meaningful qualification. If the spoken edit itself needs tightening, review that change before retiming the captions.

What if the only available video already has incorrect burned captions?

Ask for a clean source or an editable project first. If neither exists, assess whether a replacement passage or new recording is needed. Covering text or cropping the picture may remove useful information and still leave a poor result; it is not a universal repair.

About this field note

Written by Anjin Media's editorial team from hands-on work with long-form video, cut plans and searchable archives. Product details are checked against the documented platform behaviour before publication.

Read the documentation →