A listener on screen is not always an error: an interview can include an intentional reaction shot. When Descript repeatedly shows the wrong person during clear speech, however, inspect the audio-track-to-camera assignment before rewriting the transcript. The camera associated with a voice and the editorial choice to show a cutaway are different decisions.
Identify the kind of wrong cut
Mark several examples rather than diagnosing from one frame. Is the listener shown briefly during a long answer, or does every turn consistently show the opposite person? Does the error follow a particular microphone, a screen position or a sequence? Repeated reversal suggests a mapping problem; a brief reaction may match the chosen editing style.
Descript's current guidance separates active-speaker-only presentation from a style that permits cutaways. Check the selected style before calling a deliberate listener shot a failed speaker detection. Preserve that setting in the review note so another editor knows which behavior the project was actually asking for.
Trace one voice all the way to a camera
Create four columns: recorded track, audible person, intended camera and sequence position. Solo the microphone to identify the actual voice, then view the corresponding camera. A filename such as Camera 1 is not evidence that it belongs to the first person seated on screen. Shared room microphones deserve a separate note because they may contain more than one voice.
Use Camera setup to inspect the assignment. Descript determines layout positions from the order inside the sequence, not simply from how layers were arranged in the Scene editor. Rearranging the visible panels without checking the sequence can leave the underlying association unchanged. Verify both the selected camera and the person it actually shows.
A two-speaker ownership test
Imagine a fictional interview with Mei asking one short question, Luis answering alone and Mei giving a final response. The following is a proposed diagnostic, not an executed AI editing result. Keep both cameras available throughout this test interval and use a copy of the composition.
- Listen to each line with the picture hidden and name its speaker.
- View each camera without sound and identify the visible person.
- Record the audio-to-camera mapping before changing it.
- Choose a presentation mode suited to the intended test.
- Run the short sequence and review each spoken turn and its entry cut.
If a correction fixes the first turn but reverses another, check for reused or misnamed source tracks. Do not compensate by manually relabeling spoken words as the wrong person. That would make captions and later editing inherit a false identity assignment.
Keep voice ownership separate from the audio mix
The accepted picture shows the intended speaker when the selected style calls for it, and any reaction shot follows an explicit editorial choice. The soundtrack should still use its approved microphone and mix. Correct camera selection does not prove that changing angles preserved the preferred sound source.
Save the mapping with the final project and inspect a downloaded excerpt containing all three turns. This is an external Descript setup procedure, not an Imagild multicamera feature. A visual or narration brief prepared for an Imagild film can name roles, but the receiving editor still needs actual track and camera assignments. If the test contains overlapping speech, extend the review to that situation separately instead of claiming a simple three-line test validated the entire interview.