Goal
Ensure all on-screen visual information can also be represented aurally through screen narration. This benefits blind, low-vision, and non-reading players (including younger players or those with learning disabilities).Overview
Screen narration reads visible UI aloud through a synthesized voice — menu text,press A to select prompts, and important in-game info. It can be platform-level (like Windows Narrator) or game-specific. Confirm what your platform supports before deciding whether you must build your own.
Other affordances — audio cues, spatial audio, haptics (XAG 103) — are often better than narration and should also be explored.
Scoping questions
Review the categories in What should support narration. If your game contains any of them, and losing that text would block a player from configuring, starting, or completing the game, narration support is in scope.What should support narration
Support doesn’t mean narrate-by-default — it means when a player enables narration, everything listed reads aloud.- All on-screen text — menu labels, sub-labels, roles, values, descriptions
- Control interaction prompts —
Press A to select - In-game UI — HUD, inventory, objectives, hints, maps
- Player-to-player comms — incoming party chat, chat-wheel options, canned messages
- Images, diagrams, tables — via text alternatives
- Real-time updates — incoming chats, toasts, error messages, friend join/leave
When narration should fire
- Context change — opening a dialog, switching screens, entering a load screen or gameplay from a load screen
- Focus change — moving between buttons, list items, sliders (announce the newly focused item)
- Value / state change — slider values update as they change
- Real-time notifications — errors, alerts, information changes
How to narrate an item
A sighted player sees name, role, value/state, position, and interaction affordance. A player using narration should hear the same, typically in this order:- Label / name
- Control type / role (
combo box,toggle,slider) - Value / state (
Off,38%,collapsed) - Indexing (
6 of 9) — place at the end of the string. Provide a setting to disable enumeration. - Interaction model (
A to select)
Implementation guidelines
- All core UI text supports narration — main menu, options, HUD, state changes, players in the game, time-based events. Prefer the platform screen reader; if unavailable, use a speech synthesizer. Recorded audio files are a fallback but not ideal.
- Interactable elements enumerate their children and expose input type and current state / value. Example:
Worlds, Tab, 1 of 3, selectedorMusic Volume, slider, 52%. - Text alternatives for charts, diagrams, pictures, animations convey the same information the visual conveys.
- Text alternatives describe purpose and operation of UI components (e.g.
Left mouse button for show spider preview, notsymbol). - Purely decorative non-text is not spoken.
- Focus order aligns with UI meaning and operation. In non-linear layouts, focus order aligns with visual flow.
- List looping — in linear (up/down or left/right only) menus, after the last item loop back to the first. In multi-directional menus (tile grids), don’t loop. Offer a setting to enable/disable looping.
- Cancel / repeat narration must be quick, regardless of input type.
- Interrupt on focus change — in-progress narration for the previous element stops immediately when the new element gains focus; the new element then reads.
- Voice pace and pitch — allow player adjustment.
- Context change is player-initiated where possible; after context changes, announce the new context.
- Timed / recurring notifications (loading, matchmaking, countdowns) may re-read state every 7–10 seconds without interfering with other narration.
- External screen-reader support requires:
- Programmatically expose the game’s language (
en-us, etc.). If NPC dialogue switches languages, expose that too, so pronunciation is correct. - Alt text describes content — don’t include the object type in the string (
Brown shield, notImage: Brown shield). - Markup: complete start/end tags, valid nesting, no duplicate attributes, unique IDs.
- For all UI components,
nameandroleare programmatically determinable; states, properties, and values that the user can set are programmatically settable; changes are notified to assistive tech.
- Programmatically expose the game’s language (
- Time-based media has a text alternative describing the media. For live content, a descriptive title is sufficient (e.g.
Clock displaying current time in Pacific Time), not a real-time value read. - Tables are made fully accessible — column and row headers are programmatically associated with cells so context is preserved as the player navigates. Avoid a single high-level alt text on the table.
- Single cell:
Column N, row N. Column header, row header. Cell contents. - Move within row:
Column header, column N, cell contents. - Move within column:
Row header, row N, cell contents.
- Single cell:
- Pronunciation guidance — provide a mechanism for players to understand how to pronounce proper names, technical terms, or words of indeterminate language.