WCAG 1.2.9 — Audio-only (Live)
Live audio without video — a live radio show, a live audio chat, a live podcast — still has to be perceivable to deaf users. The Level AAA requirement is a real-time text alternative.
What this requires
For live audio-only content at Level AAA, a text alternative must be provided in real time. This applies to live audio streams that are not part of synchronised media — a live audio-only event, a Twitter Spaces-style room, a live audio talk page, a phone-in broadcast. The alternative is typically a live transcription stream (CART) or a real-time captioning panel displayed alongside the audio interface.
How AI coding tools fail this
When asked to "add a live audio room" or "build a live podcast page", AI tools generate the audio player and a "now live" indicator and stop. The real-time transcript stream is treated as a separate infrastructure problem — which it often is — and the UI never makes room for it.
The second pattern: an audio-only live stream embedded as <audio>
with no surrounding live region or transcript area. Even when the
platform provides a live transcription endpoint, the consuming page
doesn't wire it up.
The third: treating "we'll publish the recording later with a transcript" as an answer. That covers the prerecorded version, not the live one.
Edge cases
- Audio-only live ≠ video live. Captions on synchronised media are 1.2.4; 1.2.9 is for the audio-only case.
- Crowd-sourced live transcripts (a live community typing along) technically count if they keep up with the audio — but they're fragile. Use a dedicated captioning service for serious events.
role="log"witharia-live="polite"is the right pattern for the transcript area: new lines are announced to screen readers without interrupting the current utterance.- Latency tolerance: a few seconds of lag is acceptable; tens of seconds is not. The transcript must be usable as a real-time alternative.
- Recording for later is not the same accessibility need. The recorded version falls under prerecorded audio-only (1.2.1).
How Jeikin handles this
This is a manual review item, evaluated per live event rather than per page. The dashboard records, for each live audio event, whether a real-time transcript was provided and through which mechanism. The evidence is the event log, not an automated check.