Locate the Repetition Before Deleting It
Listen at normal speed and note the repeated words and times, then hear the individual segment. Sequential repetition may come from text, files, or placement. Simultaneous voices also require checking original sound and narration carrying the same content.
Two subtitle cues do not prove two spoken instances; one cue does not prove one mixed segment. Chosen repetition, replay, or echo can remain. Establish whether it is intentional before removal.
Repair the Layer Where It Appears
| Comparison | First repair |
|---|---|
| Submitted text repeats | Revise the current script, retaining necessary premise and reply. |
| One request, two spoken instances inside one file | Redo or replace that segment and listen. |
| The second segment uses the first file | Trace provenance and select the intended second audio. |
| Both segments are right, the export adds an instance | Check current tracks and placements. |
| Original sound and narration restate together | Revisit sound ownership rather than dropping keywords. |
The cited cache compares text, settings, and WAV size/mtime; it does not transcribe sound. Adoption checks selected order and words. Wrong words selected consistently, or correct records pointing to wrong speech, cannot be diagnosed by string checks alone.
Paper Repair: Restore the Second Window’s Words
Original plan: two Chinese segments, 0–4 seconds “她把名单折了回去。” (she folds the list back), and 4–8 seconds “门外有人敲了两下。” (someone knocks twice outside). These are illustrative windows without actual audio or proven fit. Suppose the second window repeats the first sentence.
| Layer | Hypothesis and repair |
|---|---|
| Chosen/requested text | The lines differ; preserve both and their windows. |
| Individual audio | Assume the first is correct, while an old first-line file was manually imported for the second. |
| Problem segment | Select the intended second audio; if absent, generate and hear only that line. |
| Export | Check the actual files serving both purposes, then listen through the two lines. |
Deleting the repeated 4–8-second placement alone leaves the second sentence missing. After replacement, check its ending, joins, voice, and subtitles. Changed duration requires checking adjacent positions too.
No audio, synthesis, or listening exists in this example. On the repository adoption path, match chosen text/order to current metadata and inspect consumed inputs. Matching establishes consumption, not that the recording spoke those words correctly.
Keep Repetition That Has a Purpose
A repeated line may change a reply; replay may enable comparison; echo may be chosen. Do not remove these mechanically. Give an original decision-making line room, with narration providing necessary connections. Revise ownership when two tracks unintentionally compete.
Protect unaffected files in a local repair. Avoid wiping caches or losing numbers, negation, or conditions. When a script repetition is authorized for removal, name the changed line and synchronize requests, subtitles, and chosen inputs.
Check Completeness Before the Final Listen
Review: hear the problem segment, its neighbors, then the current export at normal speed. Confirm repetition is resolved, the next line remains, and important original replies survive. Leave unlistened checks unresolved rather than inferring a pass from subtitles.
Record chosen text, request, selected file, output time, and heard content side by side. This editorial aid is not a new automatic speech-verification tool. Original dialogue cuts and missing input words need their own repairs rather than deleting one line for every fault.
FAQ
Why can audio repeat when text does not?
A recording may repeat, an import may reuse an old file, or a timeline may add an instance. Compare actual sound with current inputs; text alone cannot replace listening.
Can I simply delete the second playback?
Yes if it is an extra instance. If another line belongs there, restore it. Preserve intended replay or character repetition according to the chosen edit.
Do it with the skill
In the existing Video Recap flow, provide the repeated location, chosen text, and segment files. Ask for a minimal diagnosis using video-voiceover and downstream input records. Redo only named segments, retain unaffected text, audio, and picture, then check actual consumption and listen.
Read the method: Segment text, cache, and file records · Chosen order, text, and consumed mix inputs · Segment placement and sound ownership