kajisho5/ffmpeg-skill v3.8.22

on GitHub

Automated release: version and notes generated from pull requests merged since 3.8.21.

Fixed

• reverse.py keeps a variable-rate picture with its sound (#523). reverse gives the
  reversed frames the forward timestamps in order, so a VFR source's gaps stayed where they were
  in forward time and every flash of a 27 fps average clip came 0.19 s after its reversed beep
  (0.23 s late with --no-audio, whose picture also ran 74 ms long). The picture is now put on
  the delivered grid before the reversal, each slot taking the last frame that starts before the
  slot ends, so the reversed clip still opens on the source's last frame; the flashes land within
  half a frame of where they belong, with and without sound (FFmpeg 6.1.1). A constant-rate source
  is reversed as before.
• speedramp.py, fit.py --method speed and audio.py --tempo keep each sound where the
  retiming puts it (#523). atempo lays its first window down without retiming it and sets
  everything after it about half a window early (1024 samples at 44.1 and 48 kHz): 21 ms at 1x
  and 40 ms at 0.5x in speedramp.py, 17-25 ms in fit.py --method speed, 10-31 ms in audio.py --tempo and about 100 ms through a chain of two stages (0.3). Each chain is now led by a
  window and a half of silence and a window is cut off after it, so through one stage (0.5x to
  3x) the beeps of a test clip land within 5 ms of their place on average; through a chain below
  0.5x, where a short sound is laid down as several grains, each lands within one atempo window
  (43 ms at 48 kHz) of its place instead of 70-155 ms early. A short sound at the very start
  keeps its level (a first try with half a window faded a 10 ms tone there by 7 dB, or lost it). A 1x
  speedramp.py segment is no longer put through atempo at all. The timing.atempo stages
  audio.py reports are unchanged.
• freeze.py --at inside a gap of a variable-rate source keeps what follows with its sound
  (#523). The picture after the hold came 56 ms (--at 0.15) and 93 ms (--at 2.2) early
  against its sound on a clip with gaps from 0.1 s and 2.1 s, and the picture ran short. A hold
  inside the clip now conforms the picture as it is decoded and cuts, holds and counts on the
  delivered grid, so every frame of it is there and each flash stays within half a frame of its
  beep. A hold of one frame (--hold 0.04 at 30 fps, any source) no longer loses that frame and
  runs the rest of the picture a frame before its sound.
• Not changed: in a render.py project of several clips, --export-timeline still sizes the
  sequence from frame while the render reframes the join with fit (#526); which of the two
  wins is a decision for a minor release, with the options in docs/design-decisions.md.
• fix: variable-frame-rate sources in reverse and freeze, and atempo's early sound in speedramp, fit and audio (#541)

What's Changed

• fix: variable-frame-rate sources in reverse and freeze, and atempo's early sound in speedramp, fit and audio by @kajisho5 in #541

Full Changelog: v3.8.21...v3.8.22