The Master Guide / 10 Voice and lip sync · Speaking shots and lip sync
Checks on the 1080p result
Each Sync output passes these checks before it counts as a take. Your eye decides last.
Each Sync output passes these checks before it counts as a take. Your eye decides last.
Check | How | Pass |
Audio offset: the Sync output's audio against the plate-length file | Lay both from the same start; measure the gap on one sharp consonant | Rounded to whole frames, −1, 0 or +1. At 24 fps a frame is about 42 ms: 42 ms passes, 84 ms fails |
Frame rate and size of the 1080p finish and the Sync output | Read both files' frame rate and size (steps below) | They match the timeline, or are conformed in the edit; never regenerated for this |
Frame windows, ±2 frames, at each b, m, f and w or "oo" sound, and at the head's widest turn during the line | Find the sounds from character-level timestamps or a forced aligner; step frame by frame at 100% | Lips close on b and m, the lower lip meets the teeth on f, rounding on w; no warping, no teeth flicker, no doubled mouth |
Drift, teeth flicker, mouth warping; eyes, brows and breath moving with the line | A video-analysis report on the line's window, paid inside the approved estimate (see The quality bar → Judging motion honestly) | The report informs; it doesn't decide. If the mouth syncs while the eyes stay dead, note it and judge the shot yourself |
The words of the shipped dialogue | Transcribe it and compare with the canonical line (the approved wording; see Voice and Arabic dialects → Three texts per line) | Zero content-word mismatches |
Against the brief's references (see The quality bar → The reference bar), before you judge it | Frames beside the reference on its points | Weaker on two points: remade with a named change first |
Sync and acting at full size | You, plus the native reviewer for Saudi and Gulf lines (see Voice and Arabic dialects → The native reviewer) | Your decision |
Measuring the offset: measure the gap on one sharp consonant (the release of a t, k or b) on both waveforms, then round it to frames at 24 fps, or 25 fps for a TV version (steps below).
A "content word" is a word that carries meaning (a noun, verb, number or name), as against a small word such as "the" or "and".
Do it yourself
Measure the audio offset on one consonant (Premiere or Resolve)
In Premiere or Resolve, lay the Sync output and the plate-length file from the same start.
Zoom into one sharp consonant (the release of a t, k or b) on both waveforms and measure the gap; Premiere's audio time units step by single samples.
Use 24 fps, or 25 fps for a TV version.
If the two waveforms don't match at all, check that both files hold the same speech.
Read the frame rate and size (ffprobe)
Run this on the finish and on the Sync output, and compare the two with the timeline:
ffprobe -hide_banner -loglevel error -of default=nw=1 -show_entries stream=width,height,r_frame_rate -select_streams v:0 sync.mp4
Sources
Sync, improving lip-sync quality.