Discussions

Working on a pipeline that produces subtitle tracks and then sends them for human translation, and I want to check I am submitting them the right way before building around it.

The source SRT is machine generated first, with an AI subtitle generator giving us timings and a first-pass transcript, and then corrected by hand. So the file has accurate cue timings, but the text length shifts once translated.

Three questions:

1. Is it better to upload the .srt as a single document, or to split each cue into a separate string so translators see them as discrete segments? The docs cover file upload but not the tradeoff.
   2. Does the API preserve cue numbering and timestamps as non-translatable content automatically, or should I strip them and re-merge on my side?
   3. Is there a documented way to pass a per-segment character limit, so translations stay within a readable subtitle width?

Happy to share the merge script once I know which shape is correct.