Tutorial — Captions UI

How we show captions for everything the figure says:

Captions are on by default. Museums, schools and the ADA expect it, and many visitors are in noisy rooms.

Draft. Test with the UX and accessibility designer and with deaf and hard-of-hearing visitors before opening.

1. Standards we follow

Source	What we take
WCAG 2.2	Captions for recorded media (SC 1.2.2, Level A) and for live media (SC 1.2.4, Level AA)
FCC caption quality (47 CFR 79.1)	Four tests: accurate, synchronous, complete, well-placed (never covering faces)
DCMP Captioning Key	At most 2 lines; each caption on screen 1.33–6 s; white medium sans-serif with a shadow on a translucent box; mixed case; bottom placement; reading speed 130–160 words per minute by audience
BBC subtitle guidelines (secondary summary)	2 lines; line length up to about 68% of a 16:9 width; 160–180 words per minute
Smithsonian Guidelines for Accessible Exhibition Design	Open (always-on) captions preferred; 70% contrast; sans-serif or simple serif
Meta VR accessibility guidance	In XR: about 1 m away (or half the far-field distance) or head-locked; top or bottom of the central 40°; up to 32 characters per row, 2 rows; text scale 50–200%; background box
W3C XR Accessibility User Requirements	Support captions, and let them be customised

Setting	Kiosk (life-size display)	Headsets	Phones / web
Position	Bottom third, beside the figure, not over the face or hands	About 1 m in front, lower field, gently follows the head	Bottom, above the controls
Lines	2 max	2 max, 32 characters per row	2 max
Font	Clear sans-serif (the studio caption font), medium weight	Same, larger	Same
Size	Readable from 2.5 m (test on the actual screen)	Adjustable 50–200%	System text size respected
Colour	White or near-white on a 70–80% dark translucent box	Same	Same
Speaker label	"Maria Mitchell:" on the first caption of each turn; "You asked:" for the visitor's question	Same	Same
Sound cues	[lamp hisses], [telescope creaks] only when meaningful	Same	Same

Period spelling: caption text follows the historian-approved script exactly, including period words. Don't modernise.

Enable Subtitles and Closed Captions (Beta in 5.8).
Make a Subtitle Asset per monologue (SUB_<Figure>_Monologue_<NN>). Import the timed script as SRT, or type the entries (start offset and duration).
Add the subtitle track to the Level Sequence (Tutorial — Sequencer for Monologues), so captions stay locked to the audio.
Style the plugin's display widget (a UMG TextBlock set in Project Settings) to match section 2.
- 5.8 supports per-platform display overrides; use one for headsets.
Check timing against the audio.
- No caption under 1.33 s or over 6 s.
- Split long sentences at natural phrase breaks, never mid-name.

The conversation service sends caption chunks with timing (Tutorial — Conversation Client Plugin). They go to a UMG widget:

WBP_AT_Captions has a container with up to two lines, a speaker label and a dark translucent background.
Reveal text in phrases, in time with the audio. Don't show it word by word, which is tiring to read, and don't show it before it is spoken.
When the next phrase doesn't fit, roll up (the old line slides up and fades out over about 0.5 s).
The visitor's question shows briefly ("You asked: …") when the transcript arrives, then clears when the answer starts.
Refusals and redirects are captioned like any other line.
In XR: render the widget in a Widget Component or an OpenXR stereo layer placed per section 2. Stereo layers can draw behind the scene in 5.8, which keeps captions crisp (Tutorial — OpenXR Headset Builds).

A small accessibility menu (kiosk corner button; headset wrist menu):

Remember the choice for the session only; reset it for the next visitor.

Every monologue captioned word-for-word; timing checked
Live answers caption in sync with speech; no text before it is spoken
Never covers the face or hands at kiosk distance or in the headset
Readable at 2.5 m on the kiosk screen and in the headset (test with visitors)
Accessibility designer sign-off; notes from deaf and hard-of-hearing testers recorded