The words said in a library video or sound file, in lines with their times in the file (seconds), whether or not it is on the timeline. COST: a stretch that was not heard before is transcribed now with the user’s AI time (paid plans): one AI minute per minute of sound, silence is free. What was heard is kept, so asking again, or captioning the file once it is on the timeline, costs nothing more — the result says how many AI seconds this call used and which stretches are known. Give from / to to listen to part of a long file (get_media_levels shows where there is sound, for free). A long stretch answers with a job id: poll get_job.