Descript's duplicate-transcript or mic-bleed fix removes duplicate script text; it does not automatically mute the underlying audio. A greyed or cleaned transcript region can therefore remain audible in playback and export. Inspect the microphone tracks themselves before running the fix again. Text membership and audible signal are separate responsibilities.
Confirm what the automatic fix changed
Preserve the original sequence and compare the script before and after the fix. Then solo the relevant source track and listen at the same time range. Descript's current documentation explicitly distinguishes duplicate transcript removal from audio muting. Do not treat a cleaner script as evidence of cleaner sound.
A real user question describes this exact mismatch: the writer sees changed regions after clicking Fix but still hears bleed when soloing and exporting. That original question establishes that the confusion exists; it does not measure frequency or prove that its older replies describe today's interface.
Classify the overlap before reducing it
Identify who is actually speaking and which microphone should carry the useful voice. Distinguish the same utterance leaking into a second microphone from two people genuinely talking at once. If a quiet response is real speech, suppressing it as bleed can remove meaning.
| Region | Physical event | Separate decisions |
|---|---|---|
| One speaker, two microphones | Possible bleed | Script owner and unwanted audio contribution |
| Two actual speakers | Genuine overlap | Preserve both meanings or make an editorial cut |
| Silence plus room sound | Ambience, not duplicated words | Noise treatment and continuity |
This is a diagnostic classification, not an automatic permission to silence every secondary track.
Edit the audio with a documented scope
Suppose microphone B contains a faint copy of speaker A between B's own lines. The script fix can remove duplicate A words from the text, while B's recording still contributes the copy. On a protected duplicate, use supported sequence-level audio editing or an appropriate external audio editor to reduce only the unwanted contribution. Preserve B's real entries, breaths and required ambience.
Do not apply indiscriminate muting through a real overlap. If the wanted and unwanted sounds are inseparable in one recording, acknowledge the limitation and choose an approved compromise. Automatic enhancement is not a guarantee of speaker separation. Listen to each source and the combined mix after the change.
Approve text and sound independently
Check the cleaned transcript for missing legitimate speech, then export the reviewed interval and listen without reading. Verify speaker handovers and room continuity. Save the track/time decisions alongside the script correction. The successful outcome is an accurate script and an intentional audio mix; either can be correct while the other still needs work.
Sources and scope
Prepared by Imagild Editorial with AI assistance. The example is a proposed diagnostic, not a reported software test. These are external post-production controls; their documentation does not establish an Imagild feature.