kajisho5/ffmpeg-skill v3.8.4

on GitHub

Automated release: version and notes generated from pull requests merged since 3.8.3.

Fixed

• fit.py --method trim ends its sound to the sample, and broll.py and
  sync.py --replace-audio start a delayed sound on the sample (follow-ups of #470, the class
  of #493 and #499). fit.py --method trim --duration 1.23425 ended its output with -t 1.234:
  59232 samples for 59244 at 48 kHz, 59280 for 59261 when the millisecond rounds up
  (--duration 1.2346), 54419 for 54433 at 44.1 kHz (--duration 1.2343). The -t is
  length_output_t() now, as in overlay.py and loop.py; --method speed takes it only when
  the change is too small to retime the sound (3 s to 2.9999 s wrote 144000 for 143995), because
  a track retimed by atempo ends before the target and its -t cuts nothing (13 to 62 ms
  before it in the cases measured, which this change does not address). With sound, the same
  -t no longer lets in the frame that starts on the target: 37 frames of a 30000/1001 source
  were written as 38 under -t 1.235; without sound the command is unchanged and still writes 38.
  broll.py delayed a cutaway's sound by int(at * 1000) ms: --at 1.001 started it at sample
  48000 for 48048, --at 1.2346 at 59232 for 59261; it is counted in samples of the 48 kHz mix
  (silence_filter()), and a whole millisecond is rounded, not cut. The cutaway's picture window
  keeps its millisecond text: a --at/--from on a 30000/1001 frame whose millisecond rounds up
  still starts the picture a frame late (measured, not fixed here). sync.py --replace-audio
  rounded the measured offset to the millisecond: with --fine-ms 0.25 an offset of 0.2343 s
  became adelay=234 (the sound started at sample 11232, now 11246); at the default
  --fine-ms 1 and without --fix-drift offsets are whole milliseconds and the command is
  unchanged (--fix-drift moves the offset by the drift it measured, to a tenth of a millisecond,
  and such an offset is now a delay in samples too), and --trim-second still rounds. A length, a delay or an offset in whole milliseconds, an input without sound and
  a dry run's pending input write the command they always did, except broll.py --at values
  that were cut by one millisecond. Measured and left alone, because the millisecond cannot
  change what they write: join.py's silence for a part without sound, insert.py,
  background.py, grid.py's picture and export.py's platform maximum
  (docs/design-decisions.md, "After 3.8.3"). Measured on FFmpeg 6.1.1 only.
• fix: fit --method trim, broll's cutaway sound and sync --replace-audio's delay are written to the sample (#500)

What's Changed

• fix: fit --method trim, broll's cutaway sound and sync --replace-audio's delay are written to the sample by @kajisho5 in #500

Full Changelog: v3.8.3...v3.8.4