Health-Check Long Footage: Auto Scans for Black Frames and Silence
A three-hour stream replay or a day-long meeting recording: where did it go black, where did it go quiet — watching end-to-end is not realistic. An automated scan lists suspect segments with timestamps.
Which footage deserves a scan first
Three material types pay off most: recordings — lectures, stream replays, screen captures, where recorders occasionally drop frames into black gaps and speakers pause into long silences; meeting recordings — waiting rooms, dead air, closing small talk, with real content occupying only part; converted files — downloaded or transcoded old material with possibly corrupted middles.
The scan logic is "missing-signal detection": runs of fully black frames (blackout) and passages under a loudness threshold (silence) both get listed, each with start and end timestamps. It does not judge whether to delete — it maps where to look.
Tuning the thresholds
Two dials set sensitivity: the loudness threshold defaults to -35dB (below counts as silence) — lower it to -40 for noisy live recordings, raise to -30 for studio-quiet sources; the duration threshold defaults to 500ms (blackout or silence must exceed it to be reported) — drop to 200ms to catch blink-and-miss black flashes, raise to 2000ms to only surface "deletable dead air".
Tune wide first, then narrow: run the defaults, read the report; if dozens of entries are mostly normal pauses, tighten and rescan; if nothing reports while you clearly saw black frames, loosen and rescan. Two rounds always converge.
After the report: cut, keep, or salvage
Three actions per finding: confirmed junk (waiting, dead air, dropped-stream blackouts) — cut with the trimming tool by timestamp; confirmed intentional (scheduled breaks, black gap followed by intact content) — keep; corruption (black middle and everything after is broken) — treat as damaged footage, salvaging the intact front part if possible.
Make "scan before processing" a habit: minutes of scanning buys an accurate mental map of the footage, so every later step — trimming, compressing, merging — stands on known ground instead of discovering a broken middle halfway through.