Translate subtitles

Why subtitle files aren't just text files

Run a raw .srt through a generic text translator and it's easy to end up with a timestamp accidentally reformatted, a cue index renumbered out of order, or a line split that changes how long a caption stays on screen. None of that shows up as an error message — it shows up as a caption that appears half a second late or lingers after the actor has already moved on. translate.srt parses the cue structure first, separating timing from dialogue, so translation only ever touches what a viewer reads, never when they read it.

SRT, VTT, and what they have in common

SubRip's .srt format dates back to a DVD ripping tool from the early 2000s and became the de facto standard because of its simplicity: a cue number, a start and end timestamp, then one or more lines of text, repeated down the file. WebVTT (.vtt) grew out of that same basic idea for web video, adding support for styling and positioning cues on screen, and is now what most browsers and video players expect natively. Both formats separate timing from text in essentially the same way, which is exactly the boundary this endpoint respects — cue numbers and timestamps pass through untouched in either format, and only the spoken-line text is translated.

Line length and reading speed

Translated languages are rarely the same length as the source — Spanish dialogue commonly runs longer than the English it's translated from — and a caption that's too long to read in the time its cue allows becomes useless no matter how accurate the translation is. Text is translated with reasonable line-length conventions in mind, keeping cues readable within their existing time window rather than producing a technically correct translation nobody can finish reading before the next line appears.

Where this fits into a video pipeline

A video platform publishing in multiple languages usually generates a single source-language transcript, then needs that same file translated into every market it serves without re-timing each version by hand. Submitting the source .srt or .vtt to /translate/srt once per target language and receiving back a file with identical cue timing means the video, the audio, and the captions all stay in lockstep regardless of how many languages are layered on top. Because price is per request plus per word rather than per minute of runtime, a short product demo with sparse dialogue costs far less to translate than a dense two-hour lecture, in direct proportion to how much was actually said.