Automatic speech recognition produces words. Accessible captions convey the meaning of audio in time, including information that may never appear in a raw transcript.
What the standard actually asks for
WCAG 2.2 success criterion 1.2.2 addresses captions for prerecorded audio in synchronized media at Level A, subject to its specified exception. Criterion 1.2.4 addresses captions for live audio in synchronized media at Level AA. The W3C explanation emphasizes that captions include important non-speech information such as relevant sound effects and speaker identification, not just the dialogue. This is a content requirement, not a product certification for a particular speech model.
A practical editorial workflow
Start with an ASR draft if useful, then check each spoken word against the source. Add meaning-bearing sound cues, speaker shifts and song information where appropriate. Segment captions by syntax and comfortable reading rhythm; avoid text obscuring diagrams, subtitles or essential on-screen labels. Review both the caption file and the way the player displays it. A correctly formatted WebVTT file can still be inaccessible if the text is wrong.
Live captions require a different plan
A live event introduces delay, partial transcription revisions and audience questions. Define an escalation route for names and sensitive terms that a recognizer repeatedly mishears. Where feasible, let a human captioner correct errors, and save a corrected post-event recording. Never assert that captions were comprehensively edited if no one listened to the full source audio. Performance of one microphone setup cannot be generalized to a conference hall.
Evaluation needs separate dimensions
Word error rate is only one measure. Assess whether speakers are distinguishable, whether numbers are faithful, whether sound descriptions are included, whether captions remain on screen long enough and whether the player offers usable settings. Ask deaf and hard-of-hearing reviewers to evaluate representative clips. Distinguish a transcript, subtitles used for translation and accessibility captions in the production manifest.
Compliance scope and alternatives
This guide explains technical WCAG concepts, not which jurisdiction’s laws automatically apply to a specific organization. Regulatory duties may be different for public agencies, broadcasters and private websites; obtain region-specific legal advice where necessary. A professional captioner, a careful in-house editing process and an AI-assisted workflow can all contribute to a compliant product. The critical input is the final human-verifiable result.
Publication scope, rights and cost caveats
Research cutoff: 25 October 2026. The material above distinguishes provider documentation from independently measured behavior; no hands-on evaluation, legal clearance, voice-owner consent, commercial license purchase or model benchmark is claimed. Production teams must verify the specific API or installed version, account entitlement, input/output rights and destination market before release. Monetary plan amounts and metering tiers are not asserted here; the canonical pricing ledger belongs to AI 10. Missing release dates, exact rates and sample-rate capabilities are intentionally not inferred.