A structurally clean, synchronized, readable, delivery-ready subtitle file with a documented human review pass.
Start at step 1A subtitle file can be technically valid and still be hard to follow. Structure, synchronization, reading speed, line breaks, speaker changes, spelling, and delivery format are different quality checks. Passing one does not certify the others.
This workflow uses browser-local utilities for repeatable mechanical checks, then reserves the final decision for a human watching the exact delivery video. Keep the reviewed master untouched and create a new copy for every repair or conversion.
Each stage names the evidence to collect, the decision to record, and the tool that makes the next calculation inspectable.
Parse the complete file and repair common numbering, timestamp-arrow, spacing, and timestamp-order problems before making editorial changes.
Record the source filename, cue count, detected format, repair date, and the new output filename. Compare the first, middle, and final cues with the source.
If every cue is consistently early or late, shift the full timeline once. A growing drift or isolated bad cue needs editorial retiming rather than one global offset.
Measure the offset against a clear spoken or visual event, record the signed number of seconds, and verify cues near both the beginning and end after shifting.
Choose reading-speed and line-length thresholds appropriate to the language, audience, and program. Inspect every flagged cue for duration, line breaks, wording, and synchronization.
Record the selected characters-per-second and line-length thresholds, then list which cues require a human rewrite or retime. Do not treat an unflagged cue as automatically good.
Convert only after timing and wording are stable. SRT is a simple interchange format; WebVTT is often preferable for web players and can carry features SRT cannot represent.
Record the required destination format and compare cue count, first and final timestamps, multiline captions, speaker labels, and any positioning or styling that could be lost.
Remove cue numbers and timestamps from a copy so names, specialist terms, numbers, punctuation, and repeated text are easier to scan. Then return to the timed file for the final watch-through.
Resolve spelling and wording changes in the timed master, not only in the extracted text. Watch once muted for caption coverage and once with sound for synchronization.
Assume a reviewed 96-cue SRT belongs to an earlier cut. The editor inserts a 1.5-second intro, the web player requires WebVTT, and the team uses 20 characters per second and 42 characters per line as review thresholds.
The tools reduce mechanical errors and focus attention. Publication still depends on a complete human review of the words, timing, sound descriptions, speaker changes, and the destination player.
Use these focused guides and articles to understand the assumptions before carrying a result into a quote, production brief, or delivery file.