kajisho5/ffmpeg-skill v3.2.2 on GitHub

Automated release: version and notes generated from pull requests merged since 3.2.1.

scenes.py --beats on a music bed with no picture, and to the ms. An audio-only input
(WAV/FLAC/MP3 ...) was refused with "input has no video stream"; it is now measured as one
scene (audio_only: true), and --sheet / --shots, which need a picture, refuse it. Wrapped
in a video, a synthetic click track's grid was 40-90 ms off (sound-design-skill's 20 ms check of
generate-bgm --bpm could not pass): the 10 ms envelope is 220 samples at 22050 Hz, 9.977 ms,
and was timed as 10 ms (0.23 % stretch, 68 ms over 30 s), and the tempo was a whole number of
envelope steps (90 BPM as 0.67 s, 3.3 ms a beat). beat_grid.step_s is now the real step, the
tempo and phase are refitted to the onsets, and each onset is placed on the decoded samples
(new beat_grid.onset_refinement). On 90/120 BPM click tracks every beat is within 1 ms
(WAV/FLAC), 9 ms (MP3). cut.py --snap beats measures its own grid at the real step too.
With --beats the decode is checked (a failed decode is an ffmpeg failure, not "no steady
pulse"), and under --dry-run an input an earlier plan stage has not written is skipped
(measurement_input_pending) instead of failing the plan in ffmpeg.
Contract: scenes is audio_only: true, video_required: false; audio_only, beats and
beat_grid are described in its output schema.
fix(scenes): --beats on audio-only input, beat grid within 1 ms of a click track (#431)

What's Changed

fix(scenes): --beats on audio-only input, beat grid within 1 ms of a click track by @kajisho5 in #431

Full Changelog: v3.2.1...v3.2.2