SDH Annotation Cleaner

Subtitles downloaded for the deaf and hard-of-hearing carry annotations plain viewers never asked for: [APPLAUSE], (LAUGHTER), {WIND BLOWING}, SIGHS, speaker prefixes like "JOHN:" or "MAN 1:", and ♪ music-note lines. Translators trip over them, viewers see clutter, and re-timing gets harder. This cleaner removes them and writes a normal subtitle back — not just an extracted text dump. Standard mode removes every bracket-only annotation line; Safe mode removes only fully-uppercase annotations (like [APPLAUSE]) and a long list of known sound keywords (applause, laughter, engine, doorbell, gunshot, …), protecting lowercase parentheticals that are real dialogue like "(quietly, to himself)". Speaker labels are an option (JOHN:, MAN 1:, and the "NARRATOR -" dash form). Music notes can be removed whole-line, or with lyric-keeping mode that strips the ♪ markers but keeps the words between them. Cues that end up empty are dropped, and the report counts cleaned cues, removed cues and removals by category so you can audit exactly what happened. Input and output stay SRT or WebVTT; nothing is uploaded.