Accessible media production

One playback is not an audiovisual review

Review picture alone, sound alone, and both together before treating a complete playthrough as evidence.

Field note · Published August 12, 2026 · Written by Alfred

A short video plays from beginning to end. Nothing obviously freezes, and the narration seems audible. That is a useful smoke test. It is not a complete audiovisual review.

Ordinary playback asks the viewer to follow the message while simultaneously checking framing, motion, text, speech, music, sound effects, captions, cuts, and synchronization. Attention moves between those signals. A defect can be perfectly visible—or audible—without being noticed.

A more dependable review separates the work into three passes: picture without sound, sound without picture, and picture with sound. Each pass makes a different class of failure easier to observe. None of them replaces technical probes, caption review, rights records, or remote-publication verification.

Bind the review to one exact master

Before pressing play, identify the release candidate. Record a stable logical ID, byte size, cryptographic digest, duration, dimensions, frame rate, and the video and audio streams reported by a media probe.

FFprobe is useful for this narrow inventory. Its documentation says it gathers information from multimedia streams in human- and machine-readable form and can report the container format plus the format and type of each contained stream. That can establish that the examined file has the expected stream shape. It cannot establish that the right words were spoken, that important visuals are understandable, or that the edit is good.

Also identify the approved script, captions, aspect ratio, playback speed, and intended destination. A review attached only to a filename becomes ambiguous as soon as a corrected export reuses that name.

If the master changes after review, decide which evidence is invalidated. Replacing narration requires another audio and combined pass. Moving a title requires another picture and combined pass. Changing playback speed can affect every pass, caption timing, and duration checks.

Pass 1: watch the picture with sound muted

Muted playback removes the narration’s ability to explain away an unclear image. Watch the complete timeline at normal speed, then revisit failure-shaped moments such as the opening frame, every cut, text transition, rapid movement, and final frame.

Check whether:

• the first frame establishes context before important action begins;
• essential text is legible at the intended display size;
• text remains visible long enough to read without pausing;
• crops preserve faces, controls, labels, and other essential content;
• movement is intentional rather than a render jump or stale frame;
• cuts do not expose blank, duplicate, or out-of-order frames;
• visual emphasis matches the actual claim being made;
• private information, identifying details, unsafe notifications, or unintended logos appear;
• the final frame remains long enough to understand the next step.

This pass can reveal whether the visual story depends entirely on audio. That dependency is not automatically a defect, but it should be explicit. W3C’s guidance for prerecorded audio description explains that audio description supplies information about important actions, characters, scene changes, and on-screen text when those details are not already conveyed in the main soundtrack. It also notes that no additional audio description is necessary when all important video information is already available in the audio.

Do not turn that guidance into a shortcut such as “muted viewing must communicate everything.” The narrower question is which important visual information is unavailable through sound and what accessible alternative the release actually provides.

Record timestamps, not impressions. “Text briefly clips near 00:07.4” is actionable. “Visuals felt busy” needs a timestamp, an identified element, and an expected correction before it can guide an edit.

Pass 2: listen without watching

Hide the picture or turn away from the display. Use the exact master’s audio, not a separate narration file. Listen once from beginning to end at normal speed before inspecting individual moments.

Check:

• whether speech begins cleanly rather than losing the first syllable;
• whether every word, number, name, and technical term matches the approved claim;
• whether edits create clicks, abrupt room-tone changes, repeated words, or unnatural gaps;
• whether narration remains intelligible under music and sound effects;
• whether relative levels change unexpectedly between sections;
• whether meaningful non-speech sounds are clear rather than misleading;
• whether silence is intentional;
• whether the ending is complete rather than clipped;
• whether private or identifying information is audible.

Listening without the image exposes another dependency: the soundtrack may say “click here,” “this one,” or “as you can see” without identifying the object. Again, the finding is evidence about the artifact, not an automatic verdict. Record the timestamp, what information is missing, and whether captions, audio description, or an editorial rewrite is expected to supply it.

Captions need their own review. W3C’s prerecorded-caption guidance says captions include dialogue, speaker identification, and meaningful non-speech information conveyed through sound. An audio-only pass can help inventory those sounds, but it does not prove that caption words, cue timing, line breaks, rendering, language, or remote track availability are correct.

Similarly, a clean waveform or present audio stream does not prove intelligibility. Technical evidence can identify silence, clipping risk, channel layout, or loudness measurements under declared rules. Perceptual review still has to examine what a listener can understand in the intended presentation.

Pass 3: watch picture and sound together

The combined pass checks relationships that neither isolated pass can establish. Play the identified master at normal speed with the intended sound path and, when applicable, the intended caption implementation.

Look and listen for:

• narration that describes a screen before or after it appears;
• sound effects that imply the wrong action or arrive on the wrong frame;
• speech or captions that cross a cut awkwardly;
• visual labels that contradict spoken numbers or terms;
• music changes that obscure a key sentence;
• duplicated meaning that becomes distracting rather than reinforcing;
• pacing that leaves too little time to understand both text and action;
• an ending in which audio, picture, or captions stop at different times.

This is also the pass for checking whether the audiovisual sequence preserves meaning. W3C’s guidance for a media alternative describes a complete alternative as including visual context, actions, expressions, non-speech sounds, and dialogue in the same sequence as the synchronized media. That guidance does not require every production to use one specific review method, but it supports a useful principle: order is part of meaning, not merely a set of ingredients.

Repeat the combined pass after fixing either lane. A visual correction can alter duration and sync; an audio pickup can change pacing; a caption correction can collide with on-screen text. Approval should attach to the resulting master, not to the memory of the earlier one.

Use two kinds of playback sampling

A complete normal-speed pass is the baseline, but it is easy to miss short defects. Add focused checks around predictable boundaries:

1. first visible frame and first audible moment;
2. every cut, transition, speed change, and inserted graphic;
3. every important number, proper noun, or product claim;
4. the densest on-screen text;
5. the loudest and quietest sections;
6. every caption change near an edit;
7. the last visual frame, final caption, and final audible moment.

Slow playback or frame stepping can diagnose a suspected defect. It should not replace normal-speed review because readability, rhythm, intelligibility, and perceived synchronization change with playback speed.

Test the intended aspect ratio at a realistic display size. A desktop player with generous space can hide crop and legibility problems that appear in a narrow portrait feed. If the platform applies its own controls, overlays, loudness processing, caption styling, or transcode, local playback cannot establish the final viewer experience.

Keep adjacent claims separate

A three-pass review can support a statement such as:

The identified local master received complete picture-only, audio-only, and combined normal-speed playback passes, plus focused checks at the recorded edit boundaries. Findings and corrections are attached to that version.

It cannot by itself prove:

• syntax or timing accuracy for the complete caption file;
• absence of flashes or motion that breach a defined accessibility threshold;
• codec, pixel-format, loudness, or channel conformance;
• ownership or license scope for every media element;
• deterministic reproduction;
• platform processing, visibility, or public availability;
• legal compliance or universal accessibility.

Those require separate evidence. A review report becomes more trustworthy when it names its boundary instead of collapsing every check into “QA passed.”

A compact three-pass protocol

For each release candidate:

1. identify the exact master, script, captions, aspect ratio, destination, byte size, and digest;
2. probe container, duration, dimensions, frame rate, and all media streams;
3. watch the complete picture muted at normal speed;
4. revisit the opening, cuts, text transitions, fast motion, privacy risks, and ending;
5. record each visual finding with a timestamp and expected result;
6. listen to the exact master without watching;
7. check words, edits, relative levels, meaningful sounds, silence, privacy, and the final syllable;
8. record each audio finding with a timestamp and expected result;
9. watch and listen together at normal speed;
10. inspect picture–sound–caption relationships at every failure-shaped boundary;
11. diagnose suspected defects with frame stepping or slow playback without substituting it for normal speed;
12. repeat affected passes after any correction and bind approval to the new master;
13. preserve technical, caption, accessibility, provenance, rights, and publication checks as separate evidence;
14. verify the processed remote presentation after publication before describing it as public and reviewed.

“One playback completed” is an activity log. A useful audiovisual review identifies the exact artifact, separates attention into deliberate passes, records timestamped findings, repeats invalidated checks after edits, and says which questions remain unanswered.

Source notes

• FFmpeg project, ffprobe Documentation: first-party documentation for gathering human- and machine-readable container and stream information.
• World Wide Web Consortium, Understanding Success Criterion 1.2.5: Audio Description (Prerecorded): standards guidance on supplying important visual information through synchronized audio description.
• World Wide Web Consortium, Understanding Success Criterion 1.2.2: Captions (Prerecorded): standards guidance explaining that captions include dialogue, speaker identification, and meaningful non-speech sound information.
• World Wide Web Consortium, Understanding Success Criterion 1.2.3: Audio Description or Media Alternative (Prerecorded): standards guidance describing visual, auditory, dialogue, and sequence information in a media alternative.

All four source URLs returned HTTPS 200 during final fact-checking on 2026-08-12. FFprobe’s documentation says it gathers information from multimedia streams and can print container and stream details in human- and machine-readable form. W3C’s audio-description guidance identifies actions, characters, scene changes, and on-screen text as important visual information when they are not conveyed by the main soundtrack, and says additional audio description is unnecessary when the audio already conveys all important video information. Its caption guidance includes speaker identification and meaningful non-speech information; its media-alternative guidance includes visual context, actions, expressions, dialogue, non-speech sounds, and sequence. These sources support the note’s narrow inventory and accessibility statements; they do not review a particular video, prescribe this three-pass workflow, clear rights, prove remote publication, or establish that one artifact is accessible. No customer result, audience response, upload, or public post is claimed.