For each prerecorded synchronized time-based medium that has synchronized captions, are these captions relevant?
A deaf person watches a training video produced by your team. Captions are there, auto-generated by the platform. But proper names are truncated, a technical sentence is missing, and the silence between two lines is poorly cut. She leaves without the information. This is precisely what criterion 4.4 penalizes.
The presence of captions is not enough: this criterion evaluates their quality. Captions must transcribe all important audio information: dialogue obviously, but also sound cues that carry meaning (an alarm, laughter that signals irony, a voiceover announcing a key figure). A generic caption like "[music]" displayed while a speaker announces the conclusions of a report is non-compliance.
Synchronization is the other dimension of the criterion. Captions that arrive two seconds late break comprehension just as surely as incomplete captions. Text must appear and disappear in sync with the audio. Persistent delay? Test failed.
Un test to confirm that subtitles faithfully convey the content
Relevance and synchronization of video captions
- Identify on the page all synchronized videos (sound + image) that have a caption track.
- For each video, enable captions and watch the content with sound off.
- Verify that dialogue is transcribed in full, without significant omission or distortion.
- Verify that non-verbal audio information with meaning is indicated: alarms, laughter, speaker change, important sounds.
- Verify that text appears and disappears in sync with the audio, without systematic lag.
Test validated if conditions 3, 4 and 5 are met for each video. Test failed as soon as a caption is missing, inaccurate, or regularly out of sync.
Examples
❌ Non-compliant : Automatic caption track with omissions and errors
<video controls>
<source src="conference-accessibilite.mp4" type="video/mp4">
<!-- Track auto-generated by the platform, not reviewed -->
<track
kind="captions"
src="auto-captions.vtt"
srclang="fr"
label="French (auto)"
default>
</video>
<!--
Content of auto-captions.vtt :
WEBVTT
00:00:03.000 --> 00:00:06.500
uh criterion four point three requires
00:00:06.500 --> 00:00:10.000
that sub titles be pre-sent
00:00:12.800 --> 00:00:16.000
[inaudible]
-->Automatic captions contain parasitic hesitations ("uh"), transcription errors ("pre-sent" split at the wrong place), and mask important information behind "[inaudible]". For a deaf person, this is just as problematic as a missing track. RGAA considers that auto-generated captions not reviewed and corrected do not meet the relevance criterion.
✅ Compliant : Human caption track, complete and synchronized
<video controls>
<source src="conference-accessibilite.mp4" type="video/mp4">
<track
kind="captions"
src="captions-fr.vtt"
srclang="fr"
label="French"
default>
</video>
<!--
Content of captions-fr.vtt :
WEBVTT
00:00:03.000 --> 00:00:06.500
Criterion 4.3 of RGAA requires
00:00:06.500 --> 00:00:10.000
that captions be present for all synchronized video.
00:00:10.200 --> 00:00:12.700
[Alarm bell rings in the room]
00:00:12.800 --> 00:00:16.000
I continue: the relevance of captions, that's criterion 4.4.
-->The captions faithfully transcribe the speech, indicate the important sound event ("[Alarm bell rings in the room]"), and are synchronized to within half a second. A deaf person receives exactly the same information as a hearing person.
Tips and pitfalls
⚠️ Automatic captions never validate criterion 4.4
YouTube, Vimeo and most platforms generate captions by AI. These tracks systematically contain errors: omissions, word confusion, lack of punctuation, lag. This is the most frequent error found in multimedia audits. For an automatic track to be acceptable, it must be exported, fully reviewed and corrected before being reimported. Human work is non-negotiable.
⚠️ Forgetting non-verbal audio information
Captions are not just about dialogue. An alarm that signals danger, applause that marks the end of an intervention, a crash sound off-screen mentioned later on screen: if these elements convey information, they must be transcribed. The usual convention is a description in brackets: [Applause], [Fire alarm], [Laughter in the room].
💡 Audit by sampling on long videos
For a 45-minute video, watching it all is not always realistic. Take at least three representative excerpts: beginning, middle, end. Focus on passages with technical terms, numbers, proper names or speaker changes. These are the areas where relevance errors concentrate.
⚠️ Multiple speakers in the same video
When two or more speakers talk, captions must allow identifying who is speaking. RGAA does not impose a specific format, but common practice is to indicate the name or role before the line: "Moderator: What is your position?". Without this identification, a deaf person can lose track of a multi-voice debate.
⚠️ This criterion only applies to videos that already have captions
Criterion 4.4 only checks the relevance of captions for videos that already have them. If a video has no captions, it's criterion 4.3 that applies (existence of captions). The two criteria are distinct and must be evaluated separately: 4.3 checks that they exist, 4.4 checks that they are correct.
Frequently asked questions
How are automatic captions from YouTube or Vimeo evaluated according to RGAA 4.4?
No. Automatic captions do not satisfy criterion 4.4 as-is. The only way to use them is to export them to VTT or SRT format, correct them fully (transcription, punctuation, synchronization), then reimport them as a manual track. Without this human review step, they cannot be considered relevant.
What is the difference between RGAA criterion 4.3 and criterion 4.4?
Criterion 4.3 verifies that synchronized captions exist for videos that need them. Criterion 4.4 verifies that these captions are complete and correctly synchronized. A video can pass 4.3 (captions are present) and fail 4.4 (they are inaccurate or out of sync). The two criteria are evaluated separately.
How to concretely audit the relevance of captions according to RGAA criterion 4.4?
Enable captions, mute the sound, and watch the video. If you understand all the content thanks to the captions alone, they are relevant. If you feel gaps, confusion or lose track at certain moments, they are not. This comprehension test without audio is the simplest and most reliable operational criterion.
When should background music and ambient sound appear in RGAA captions?
Only if they convey information. Purely decorative background music does not need to be indicated. On the other hand, if music changes to signal a structuring emotional passage, or if an outside sound is mentioned in the speech, it must be transcribed. The rule: caption what a hearing person perceives and what modifies their understanding of the content.
What level of synchronization lag causes failure on RGAA criterion 4.4?
A one-time lag of less than one second is generally acceptable. It's a systematic or recurring lag that constitutes a failure: if text regularly arrives late or early, reading becomes difficult for a deaf person trying to link the speaker's gestures to the displayed words. Test on at least three representative excerpts before concluding.