How Should Captions Identify Two Offscreen Voices in a Product Advertisement?

Determine when speaker identity changes meaning

Captions convey more than the words spoken. W3C guidance on prerecorded captions explains that relevant audio information includes speaker identification and meaningful non-speech sounds.

In a product advertisement, two offscreen voices might represent a demonstrator and a narrator. If one says 'Press the side button' and the other asks 'Does that release the lid?', the distinction helps the viewer understand the exchange. Without identification, the captions may read as one person contradicting or questioning themselves.

Start from the final mixed audio. List each speaking role, the intervals when that role speaks and the approved words. Do not infer an actual customer's identity from a staged dialogue or label a generated voice as a verified testimonial.

If the voices have no meaningful distinction and the dialogue remains understandable without labels, assess the actual context rather than mechanically repeating a name on every cue. The purpose is attribution where it is needed.

Choose stable, truthful labels

Use labels such as 'Narrator' and 'Demonstrator' when those roles accurately describe the production. Use a person's name only when it is approved and appropriate. Avoid alternating among 'Host,' 'Presenter' and 'Speaker 1' for the same role without a reason.

Inspect whether the visual edit ever identifies a speaker. A person appearing on screen does not necessarily establish that the voice belongs to them, particularly in AI-assisted or dubbed content. Keep the caption assignment tied to the audio production record.

For each cue, compare the speaker label, spoken words and time interval. If a line begins before a camera cut, its speaker identity should not change merely because the picture changes. If a narrator reads across several scenes, preserve that continuity.

Caption formats and players differ in how speaker information is styled. Choose an implementation supported by the destination and verify the actual rendered result. A worksheet label alone is not a published caption.

Hypothetical case: a question is attributed to the wrong role

Hypothetical illustration; no audio or caption file was tested. A fictional lamp demonstration uses two offscreen roles. At 1.0–2.5 seconds, the Demonstrator says, 'The switch is on the base.' At 2.6–3.5 seconds, the Narrator asks, 'Can I see it closer?'

In proposed caption version A, both cues are labeled Narrator. The words may be correct, but the attribution is wrong under the stipulated speaker map. In proposed version B, the first cue identifies Demonstrator and the second identifies Narrator.

The correction should be verified against the actual final recording. If the audio production later replaces the second voice with the demonstrator's voice, the caption map must be revised rather than retaining a fictional two-person exchange.

Review the finished captions without relying on sound

Play the final video with captions visible and sound muted. Ask whether the reader can follow who gives instructions, who asks questions and whether the dialogue changes roles. Then listen with sound to verify that the labels match the actual voices.

Check product names, units and exclusions independently. Correct speaker attribution does not compensate for an incorrect transcript. Also verify that labels do not crowd the caption area or obscure a relevant product action.

The existing automatic-caption proofing guide addresses transcription accuracy. This guide adds the distinct responsibility of mapping offscreen voices to their captioned speech.

Record your own observations

The filled rows are illustrative examples. Use the blank row for your own files and observations.

Download the editable CSV worksheet

What the references establish

• W3C WAI — Understanding Captions (Prerecorded): Primary accessibility guidance explaining that captions include speaker identification when necessary for understanding.
• W3C WAI — Captions: Easy Checks: Provides practical checks for identifying speakers when multiple people speak and verifying caption synchronization.