Check Source Audio Before Moving a Cut
Mute music and narration temporarily, listen to source-audio joins, and compare the same line in the source.
| Symptom | Decision and repair |
|---|---|
| Source contains the ending; the edit leaves mid-line | Move the planned out point to retain the full utterance and needed pause |
| Ending is present but masked by music or narration | Adjust that track or narration placement; a picture cut cannot fix masking |
| Ending is missing in the source too | Check the selected audio stream or another complete take; otherwise record or rewrite rather than claim restoration |
| Only automatic captions omit a word | Confirm complete sound, then correct text/timing without recutting a good audio track |
A waveform dip does not establish a finished sentence: a quiet syllable or brief pause may be part of speech. Transcription helps locate the line; actual listening decides completeness. This article addresses the first case: a new cut truncating intact source dialogue.
Original Timing Example: The Missing Word Changes Permission
This is a paper edit with illustrative times, not analyzed or played footage. In a book-borrowing scene, source 10.20–12.60 contains “You can borrow it, but return it before Friday.” Known pauses are 10.00–10.20 before and 12.60–12.90 after. The current 10.40–12.30 selection loses the beginning and part of the return condition.
A proposed 10.10–12.75 selection puts both edges inside known pauses and retains the line. This is no fixed “add half a second” rule: choose from actual speech, pauses, and actions, without cutting into the next speaker.
| Plan | Timing record |
|---|---|
| Old segment | Source 10.40–12.30, duration 1.90; with 5.00 seconds preceding it, output 5.00–6.90 |
| Repaired segment | Source 10.10–12.75, duration 2.65; output 5.00–7.65 |
| Complete line | Source 10.20–12.60; at unchanged speed, new output 5.10–7.50 |
| Following segment | Previously started at output 6.90; now 7.65, moving 0.75 seconds later |
These are assumed times. Use actual rendered duration and frame boundaries for real footage rather than treating worksheet arithmetic as measured output. Listen for the whole permission and return condition; complete caption text alone does not prove complete sound.
Recheck Speech After Visual Boundary Cleanup
video-cut cleans edges near visual hard cuts first, then adjusts to reliable sentence or quiet windows, with speech completeness checked last. Sentence anchors use medium/high confidence records. An edge still inside known speech becomes a blocking issue rather than a successful unsnapped cut.
Without usable speech timing, the implementation marks the relevant check unchecked; it cannot automatically promise safe dialogue. Listen and obtain reliable timing before choosing boundaries. Pause windows also cannot establish meaning: keeping words still requires retaining an earlier “may” or “must not” where relevant.
Truly adjacent intervals from the same source, continuous in both source and output time, have not removed speech merely because they have separate clip labels. Listen to that join too. Different source files need their own speech intervals, not copied pause times. If visual cleanup leaves half a spoken line, repair selection again; zero flash-frame candidates is no audio approval.
Update Later Sound and Captions After Extending a Line
In the example, later material moves 0.75 seconds. Narration, captions, and effects written for 6.90 cannot be copied unchanged. Source selections retain source times; edited narration uses new output times. Bind captions to the current video and chosen sound, then check starts, final words, and reading time.
Do not shift all captions by 0.75: the preceding five seconds stay unchanged, and the repaired line’s relative position changes with the earlier in point. Correct affected intervals and later cumulative times, then review the passage. Speed changes, reordering, or frame rounding require actual output measurements rather than one offset.
Play the exported join normally and listen without picture for complete speech, duplicated words, or an abrupt fade. With picture, confirm actions and responses belong to that line. If source material is inadequate, leave recording or selection unresolved instead of claiming that a caption, freeze, or transition restored audio.
FAQ
Can a fade restore missing words?
No. A fade can smooth a join within a safe pause, but cannot generate a removed syllable. Restore complete source sound first, then treat the join if needed without fading away speech.
Can picture cut away while dialogue continues?
You can edit picture independently while retaining complete sound, using a reaction or insert that belongs to the event. Moving coupled audio/video boundaries is not the same operation; check actual correspondence in a separate-track edit.
Do it with the skill
Use video-recap to diagnose the specified clipped line against source audio. Distinguish a bad cut, masking, and source loss; propose boundaries around the complete line, check pauses and speech spans, and update later narration/captions to actual edited times. Provide a normal-speed listening candidate.
Read the method: Cut boundaries and speech priority · Pause windows, speech spans, and continuous joins · Line review in source and finished audio