raw-video-processing

Clean up raw screen recordings before publishing by removing dead air and tightening the pacing. Screencasts tend to accumulate long pauses while thinking or waiting for things to load, plus low-level keyboard and background noise, and they often feel a touch slow. This skill addresses all three with two FFmpeg-based Python scripts run through uv, and FFmpeg and uv are the only prerequisites.

By default both scripts run in sequence. First, remove_silence.py detects and cuts silent portions, invoked with an aggressive `-t="-20dB"` threshold, where the `=` syntax avoids argparse misreading the negative value as a flag, and `-d 0.5` to remove even short silences, producing an `<input>_nosilence.mp4`. It prints a summary of the segments found, total silence removed, and the kept segments with timestamps so the result can be sanity-checked. Second, speed_video.py applies a playback multiplier, `--speed 1.2` by default as a subtle boost that does not feel rushed, producing `<input>_nosilence_1.2x.mp4`.

Order matters: silence removal must come before the speed pass, because detection works on the original audio and speeding up first would distort the audio and reduce accuracy. Both scripts expose options for custom output paths, and each supports tuning, such as conservative cleanup with `-t="-30dB" -d 0.8`, extra-aggressive cleanup with `-t="-15dB" -d 0.3` plus `--speed 1.5`, or a `--dry-run` preview that lists what would be cut without writing a file. Quality is preserved throughout, since remove_silence uses stream copy with no re-encoding while speed_video re-encodes with FFmpeg defaults. Long videos over 30 minutes may take a few minutes to process.