More context than dialogue alone
SDH means subtitles for deaf or hard-of-hearing viewers. Alongside spoken words, they can identify speakers and describe sound that matters to the story. A speech-to-text draft alone does not contain all of that information.
Add information that changes understanding
Use concise labels when a speaker is not visually clear, and descriptions such as [doorbell rings] when the sound affects the scene. Music descriptions should convey relevant mood or information rather than transcribe every background noise.
Review the whole scene
Watch without sound to see what information is missing. Check that labels do not cover important action and that viewers have enough time to read them. Automated subtitles need human editing before being described as a finished SDH track.