Include more than the lyrics
Identify a speaker when the identity is needed to understand the scene. Include meaningful audio cues where appropriate, such as a bell that marks a transition. Keep captions synchronized with the recording. A transcript is useful for reading separately, but it does not replace timed captions during playback.
Review automatic output
Names, regional vocabulary and sung words can be difficult for automatic systems. A human review should compare the text against the actual recording and correct timing as well as spelling. If a passage cannot be understood reliably, acknowledge the uncertainty instead of inserting a plausible phrase that changes the meaning.
Give the viewer control
Use a player whose caption controls are easy to find. Make the text readable without covering essential visual information. Where translations are offered, identify the language accurately and preserve access to the original. Accessibility is part of the listening experience, not an extra feature reserved for a separate audience.
