Confirm Which File You Are Hearing
Open the adopted version and check player volume, selected track, and output device against a known audible file. An old preview, picture-only candidate, or another track is not the delivery. If the reference is also silent, address playback first.
Listen to the intended track in a multi-track file. An ordinal is a position, not proof track zero contains dialogue. Record file and time; opening it or finding an audio field proves no audible target sound.
Find Where Sound First Disappears
| Comparison | Check |
|---|---|
| Source | Intended stream and actually recorded sound at the selected interval |
| Edited intermediate | Consumed source and interval; silence padding for a source without audio |
| Segment WAV or adopted full mix | Existing, complete content actually consumed this run |
| Current video | Actual mode, selected stream, and attenuation or masking |
Video Recap queries streams and checks the probe process result. Failed probing and successful confirmation of no target stream are different observations; inspect errors before declaring no recording. The ffprobe documentation describes stream inspection and errors when input cannot be opened. A detected stream still needs playback.
Gain cannot recover unrecorded speech. Repair selection or routing if the source is complete but the intermediate loses it. Address a missing or unplaced WAV locally rather than adding music to hide it.
Paper Example: One Silent Source Is Not a Silent Whole
These are hypothetical sources and checks, without rendering or listening. A supplies three seconds with recorded dialogue; B supplies two with no audio stream; C supplies three with ambience. A–B–C totals eight seconds.
| Output | Expectation and pending check |
|---|---|
| 0–3 | A’s dialogue; listen to its selected source interval and intermediate. |
| 3–5 | B’s equal-length silence padding; unrecorded sound does not appear automatically. |
| 5–8 | C’s ambience; B being silent should not discard C. |
The cited multi-source renderer handles each source separately: trim existing audio, or create equal-length silence, then concatenate. A continuous audio track can contain a silent passage.
If all 0–8 is actually silent, B alone is not the explanation. Compare A/C source audio, the intermediate, and final mode. For B alone, check available source sound or intended silence. Newly designed sound is a separate decision, not recovered recording.
Check the Actual Mode, Then Hear the Delivery
Narration needs valid placed segments. Source-mix creates no narration but processes the selected source audio. Adopted AAC copy has codec, interval, and input constraints; a wrong selected stream copies wrong content. Choose no-narration explicitly rather than using empty TTS as completed speech.
A copy failure does not authorize a silent switch to mixing. Pair new picture with an old complete mix explicitly, then check output selection. Changing modes may change processing; a renamed mode proves no retained mix.
After repair, listen to the target interval, joins, and tail, then watch the full video. Record which layer was repaired and what remains unheard. Stream inspection, packet copying, and playback prove different things.
Complete repair: replace the silent preview and retain source sound
Keep A’s three seconds, B’s two and C’s three. This branch assumes checks establish that audio ordinal 0 in edited_source.mp4 already contains A’s complete dialogue, B’s equal-duration silence and C’s ambience. A picture-only preview was mistakenly delivered. The selected version uses source sound, without narration, extra BGM or new captions.
Complete repair request:
Use the checked eight-second cut as input, retaining A, B’s silence and C. Generate no narration or added music. Verify copy-compatible audio and picture clocks, then explicitly adopt adopted-packet-copy with audio ordinal 0. Use new work/delivery directories and retain the old preview. Open the new delivery, listen to all three intervals, both joins and eight seconds, and compare picture order.
From the installed video-assemble directory, the paper command is:
python3 scripts/assemble.py /project/edited_source.mp4 --work-dir no-sound-repair --output-dir no-sound-delivery --recap-stem no-sound-candidate --audio-mode adopted-packet-copy --audio-stream-index 0 --no-burn-subtitles
Replace the path and use new directories. Copy requires selected AAC and compatible intervals, does not read TTS and rejects extra BGM or explicit TTS settings. No new burn-in follows this example’s caption choice; source-burned pixels remain. Choose any new captions separately; see missing-caption diagnosis.
Copy applies to the actual cut’s audio, not untouched original AAC packets from A and C: multi-source cutting has already processed and joined sound. If the cut lacks A or C, repair it before freezing the audio.
No execution or listening occurred here. B is deliberately silent; review A and C for complete audibility. Music in B does not recover the whole soundtrack.
FAQ
Why can a file have an audio stream yet sound silent?
It may contain silence or different content, or playback/mix settings may affect it. Listen to the selected track and compare source with output.
Does one silent source silence every clip?
The cited multi-source implementation pads each missing-audio segment separately. Other audio should remain; a fully silent result still needs interval, probe, and mode checks.
Do it with the skill
Ask video-recap to diagnose this specified voiceover or assembly problem. Identify current media and time, observations, missing evidence, and a limited repair. Preserve approved text and unaffected sound and picture; mark ungenerated or unheard parts unchecked rather than treating metadata as listening approval.
Read the method: Audio-stream probe conditions · Per-source sound and silence padding · Source, narration, and copied audio modes · Missing WAVs and placement state · Assembly command and caption choice · Cut audio and new-directory routing