Non-speech sound needs description, not only dialogue transcription

What it is

Non-speech sound has to be described separately. A door, an alarm, footsteps, a musical theme—these are not dialogue and so never appear in a captions track that transcribes speech alone, yet they often carry the plot turn, the location cue, or the emotional setup. Without them, a reader gets a world in which only people talk.

Why it happens

Non-speech audio does two jobs: event signalling and context building. The first includes alert tones, alarms, and cue effects that directly drive what the user does next; the second includes ambience and score that shape emotional reading. Transcribing dialogue skips both, because neither has a linguistic form. Filling the gap requires someone—person or reliable model—to identify the sound's meaning and render it as text, which cannot be derived from the dialogue itself and therefore needs its own strategy and effort budget.

Studying it

Have description-only readers restate what happened, who was present, and whether a danger cue occurred, compared against audio-only and full audiovisual viewers, measuring event identification and context misreading. Variables include description granularity, whether event and atmosphere sounds are treated separately, and conflicts with the dialogue timeline. Separate missed events from misjudged atmosphere, since they break comprehension differently.

Where it stops holding

Not every sound needs description. Continuous background noise or decorative effects that do not affect comprehension will only drown the dialogue if transcribed. Description is also bounded by the timeline: sounds overlapping speech must fit into narrow gaps, forcing priority choices. If a sound already has a visual equivalent—a visible door closing—an extra line may be redundant.

Applying it

• First enumerate the non-speech sounds that affect comprehension and action, separating event from atmosphere.
• Describe sound and source in short present-tense phrases—"a siren in the distance"—rather than listing acoustic parameters.
• When description and dialogue compete for the same slot, keep the event sound.
• Verification: give users only the description text and have them restate events and setting, checking whether any sound that affects later plot or action was missed.

Related

• Within the group: D2.12.1 Captions carry the words but rarely restore tone or urgency · D2.12.4 Equivalence is not full replacement; some sound information is lost in transcription
• Adjacent: J2.03 Coverage of captions and text alternatives · D2.01.1 Abstract alert tones must be learned to build a mapping
• Search terms: audio description · non-speech cues · captions