Why Your Subtitles Drift, and the Arithmetic That Fixes It

When video subtitles do not line up with the spoken dialogue, diagnosing the nature of the error is the necessary first step. Subtitle timing errors fall into two distinct mechanical categories:

1. Constant Offset (Shift): Subtitles that are wrong by the exact same amount all the way through require a linear shift. If the first line is exactly 1,200 ms early and the final line is also 1,200 ms early, adding a uniform offset of +1.200 seconds across every cue resolves the error completely.
2. Progressive Drift (Rate Error): Subtitles that start about right and get further out as the film runs are wrong by a rate, and no uniform shift can fix that. If you shift the file so the beginning matches the dialogue, the ending will drift noticeably out of sync. Conversely, adjusting the end throws off the beginning.

The usual cause of progressive subtitle drift is a frame-rate mismatch between the video master used to time the original text track and the video release being played. Common standard production frame rates include 23.976, 24, 25, 29.97, and 30 fps.

When a subtitle file authored for a 25 fps PAL television broadcast is played against a 23.976 fps NTSC film transfer, the dialogue in the film runs slower than the subtitle cue timestamps. The timing ratio between the two standards is:

$$25 \div 23.976 \approx 1.042709$$

Each second of video duration takes approximately 1.0427 seconds of elapsed playback on the 23.976 fps release. While a 4.27% difference appears modest in a five-second scene, the discrepancy compounds across feature-length content:

• Over a 10-minute short: $600\text{ s} \times 0.0427 \approx 25.6\text{ seconds}$ of drift.
• Over a two-hour film (7,200 seconds): $7,200\text{ s} \times (1.042709 - 1) \approx 307.5\text{ seconds}$, which is about 5 minutes of cumulative drift.

By the second hour, subtitles appear several minutes ahead of the corresponding audio track.

Rather than guessing unknown historical frame rates or intermediate conversions, the mathematically sound fix is two-point synchronization.

Two-point synchronization works by anchoring two known reference points: 1. First Line Reference: Note the true audio time when the first spoken subtitle line occurs ($T_{\text{actual}, 1}$) versus where the file currently places it ($T_{\text{file}, 1}$). 2. Last Line Reference: Note the true audio time when the final spoken line occurs ($T_{\text{actual}, 2}$) versus its timestamp in the file ($T_{\text{file}, 2}$).

From these two data points, we calculate a global scale factor ($S$) and a global initial offset ($O$):

$$S = \frac{T_{\text{actual}, 2} - T_{\text{actual}, 1}}{T_{\text{file}, 2} - T_{\text{file}, 1}}$$

$$O = T_{\text{actual}, 1} - (S \times T_{\text{file}, 1})$$

Every intermediate timestamp $t$ across the entire subtitle file is then recalculated using a single linear transform:

$$t_{\text{adjusted}} = (S \times t) + O$$

With just two recorded numbers, the entire subtitle track stretches and shifts proportionally, landing every spoken line in between on its exact dialogue mark.