Correcting a transcript can change only the displayed words while leaving the recording untouched. If the speaker genuinely said the wrong word, a spelling correction does not replace that speech. First decide whether the recording is correct and the transcript is wrong, or whether the spoken content needs an authorized new take. These are different tasks with different acceptance checks.
Establish what was actually spoken
Listen to the original phrase without looking at the transcript, then compare it with the approved wording. Mark uncertainty rather than letting the existing text lead your hearing. A name written incorrectly may simply be a recognition error; a wrong date clearly spoken in the recording requires a different editorial decision.
Save the source, the first transcription and a working copy. Record which instances are wrong. A correction that is appropriate for one person's name may not be appropriate for a similar ordinary word elsewhere in the interview.
Use the correct editing mode
Descript's Correct mode documentation explicitly separates transcript correction from underlying audio and video. Its individual and all-instance options also have different scope. Verify the mode and intended scope before a bulk correction, then review the occurrences rather than assuming every matching spelling means the same thing.
If the source words themselves need replacing, obtain an approved new recording or use a suitable authorized workflow in the external editor. Descript's recorder guidance distinguishes inserting, replacing a selection and adding a separate layer. Those are Descript capabilities; they do not establish an equivalent Imagild speech-editing tool.
A hypothetical date correction
Imagine the speaker says “October fifteenth,” while the transcript displays “October fifth.” Correct the transcript and replay the same phrase: the audio should still contain the original correct date. Now consider a second clip in which the speaker really says the wrong date. Changing its text alone would produce disagreement between caption and sound.
This is a hypothetical two-case test. Use a table with source speech, approved text, chosen action and playback result. If a new take is used, label it as a new take and review its entry, exit and picture relationship. Do not describe newly generated or newly recorded speech as recovered original material.
Verify the delivered combination
Check text accuracy, audio accuracy and synchronization independently. Wait for any re-alignment process to finish before making further text-based cuts, then audition the neighboring words. A correct visible sentence does not prove that a later cut used correct word boundaries.
Accept when the transcript matches verified speech, or when an approved replacement is reflected accurately in both audio and captions. Reject a caption that silently contradicts the recording, an unintended global name change or a replacement that truncates surrounding words. Keep the correction log with the final transcript. Imagild's Skills can help with visual development; verification of the actual spoken material remains part of the chosen external editing and approval process.