Bypasses transcription by feeding the pipeline a pre-built `Document` whose words come from the textarea (durations from a char-count heuristic). The default segment and line splitters reformat the input into nicely-sized captions.