How do you proof automatic captions for a product video?

Start from the final audio

For an illustrative narrated pouch clip, a speech recognizer might turn a model name into a common word or confuse “does not include” with “does include.” Listen to the final edit while reading its captions. Do not proof captions against an earlier voiceover that was replaced during editing.

Current Itemlume product clips do not contain generated audio. Narration and captioning happen in external tools. W3C explains that captions represent speech and relevant non-speech audio; a promotional text overlay has a different role.

Use a high-risk token list

Keep the list short and specific to the clip. The point is to direct attention to errors that affect product understanding, not to replace a full listen.

Inspect timing and reading

Play at normal speed on a phone. Check that captions appear with the corresponding speech, remain long enough to read and do not cover the relevant product action. Correct punctuation where it helps comprehension without changing the speaker’s meaning.

Approve the exported version

Inspect the exact file or caption track uploaded to the destination. Verify the final audio and subtitle language selection together. Keep the corrected caption file with its edit version. Read track versus overlay choices and audio claim review for related decisions.

A negation error example

An illustrative final narration says that a stand is not included, while the automatic caption omits the word “not.” Reject the caption despite the rest of the transcription being fluent. Correct it against the final audio, then watch the exact export at normal speed to verify timing. Check whether another edit replaced the narration after caption approval. Keep the corrected track with the clip version. A single missing word can reverse the order promise, which is why product captions need factual proofing as well as spelling review.