Caption QA Studio

Checking captions before they ship

Caption quality is measured against how fast somebody can read, not against whether the words are correct. A perfectly transcribed caption flashing past in half the time a reader needs has failed at the only job it had. Most caption problems are timing problems wearing a spelling costume.

Reading speed is the primary check. Broadcast practice generally holds captions to somewhere around 160 to 180 words per minute for adult audiences, and lower for children's programming. A cue exceeding that is not readable regardless of how accurate its text is, and the fix is either holding it longer or splitting it in two.

Line length and line count come next. Two lines is the working maximum for most delivery, and around 37 to 42 characters per line is the usual ceiling. Three-line cues push the picture out of view and force the eye to travel further than a glance allows. Where a cue must break, breaking at a grammatical boundary reads far better than breaking at whatever character the limit falls on.

Minimum duration matters as much as maximum. A cue on screen for less than about one second registers as a flash rather than as text, even when it holds only two words. Setting a floor and merging anything below it removes a whole class of complaint.

Gaps between cues need to exist. Captions running back to back with no frame between them make it impossible to see that the text changed, so a reader loses their place. Two frames of gap is a common minimum and costs nothing.

Speaker identification and sound cues are what make captions usable rather than merely present. A conversation with no indication of who is speaking is guesswork, and a scene whose tension comes from a sound nobody has labelled loses that tension entirely for anybody relying on the text.

Position matters where the picture carries information. A caption sitting over a name on screen, a subtitle already burned into the source, or a speaker's mouth removes something the viewer needed. Most caption formats allow a cue to be moved for exactly these moments, and the ones that do not are a reason to prefer a format that does.

Check the file rather than the preview. A player renders captions using its own rules for line breaking, timing tolerance and character limits, so a file looking correct in one preview can break in another. Validating against the numbers means the result is predictable everywhere rather than confirmed in one place.

Delivery profiles differ, and checking against the wrong one wastes the pass. A broadcaster's specification, a platform's requirements and a regulator's accessibility rules all set different limits on the same properties. Confirm which profile applies before running a check, because passing the wrong standard is indistinguishable from failing the right one.

Vocabulary

Cue. One caption unit, with its own start time, end time and text. The basic building block of a caption file, and the thing every quality check is applied to individually.

Reading speed. How fast caption text arrives, in words or characters per minute. The single most important caption quality measure, because text nobody can finish reading conveys nothing regardless of accuracy.

SRT and VTT. The two common caption file formats. SRT is older, simpler and very widely supported. VTT is used on the web and supports styling and positioning that SRT cannot express.

Delivery profile. The set of caption rules a particular broadcaster, platform or regulator requires, covering reading speed, line length and timing. Different profiles set different limits, so a file passing one can fail another.