Skip to content

Industry field guide

Postproduction caption and cut alignment data for AI

A caption can be accurate for one cut and drift after a new opening is added.

2022 NAICS
512191
Updated

What this guide tests

Which cut does CA directly support, and what timing difference follows for the line in Cut B without implying a text error?

See the specimen and fields · Work through the fictional records

Task and source context

A caption can be accurate for one cut and drift after a new opening is added. EBU’s audiovisual metadata work provides an asset and media-resource vocabulary. A postproduction team can show a prospective buyer a small, cleared alignment exercise with the source cut and output clearly identified. EBU media resource metadata

Create an original clip and two edit variants with a different opening length. Give the reviewer a cue file, source cut ID, timebase, and one mismatched cue. Ask which export the cue supports and what is needed to align the other. Keep spoken text and timing errors as separate labels.

Clear the script, performance, recording and derived text separately. Exclude client edits, works in progress, embedded music and protected source files from the first sample.

Describe a caption alignment dataset

Industry research

The U.S. Census Bureau’s industry definitions establish which activities these codes cover. The workflow sources in this guide describe the task or record type. Fiund’s pilot below uses fictional material to test a specific decision within that scope.

512191 · Teleproduction and Other Postproduction Services

This U.S. industry comprises establishments primarily engaged in providing specialized motion picture or video postproduction services, such as editing, film/tape transfers, subtitling, credits, closed captioning, and animation and special effects.

U.S. Census Bureau, 2022 NAICS Manual, p. 404 ↗

Build a pilot sample

Set up this pilot

Use an original caption line and two edits with different opening offsets. Convert cue timing on the stated timebase while grading text accuracy independently.

Prepare the test packets

  1. Start with the worked case. Use the fictional record types below as a guide to the source files you need. Give each packet a stable ID and save its source versions with it.
  2. Change one fact. Make one packet with enough evidence to answer the question and one where the evidence points to a different answer. Keep the other fields fixed so a reviewer can tell what changed.
  3. Remove the decisive evidence. Give the reviewer a third packet with a plausible near-match in its place. Have them write what can still be established and what evidence would resolve the rest.

Fields the reviewer needs

Cut and cue-file identity
Defines the timeline a cue targets.
Timebase and opening offset
Makes the simple alignment calculation reproducible.
Text versus timing label
Separates two kinds of evaluation error.

Use Make a segment-level rights inventory for media and publishing to turn the specimen into a brief with a defined task, source and review owner.

Fictional exercise

An added opening shifts cues without changing their words

Dummy edit timelines describe an original short script. Time is measured in seconds from each cut's start, and no frame-rate conversion is required.

Records supplied to the evaluator
Record IDSourceWhat it records
PT1Cut A and cue file CASpoken line Harbor at 5.0–6.0 seconds; CA explicitly targets Cut A on its seconds timeline.
PT2Cut B edit mapAdds exactly 2.0 seconds before all Cut A content; no other edits or speed changes.
PT3Caption reviewCA words match the original script. Cut B export still uses the unchanged CA cue at 5.0–6.0; no retimed file supplied.

Question for the model. Which cut does CA directly support, and what timing difference follows for the line in Cut B without implying a text error?

Show the reference answer

PT1 directly supports CA's 5.0–6.0-second cue for Cut A. PT2's explicit two-second opening moves the same line to 7.0–8.0 in Cut B. PT3 confirms the words but shows the unchanged cue, so the documented issue is alignment rather than transcription. Preserve source-cut identity and the edit map when proposing a retiming review; no corrected export has yet been recorded.

Download the fictional records and answer key (JSON)

How to check the result

Ask for the answer, the record IDs that support it and any fact the evidence leaves unresolved. Have a reviewer compare the response against the reference before deciding whether the sample is ready to expand.

Errors to flag in this case

  • Calling correct caption words wrong because their timing belongs to another cut.
  • Reporting that a retimed export exists when only the required offset has been calculated.

Change the inputs and run it again

Replace the uniform opening with a mid-cut insertion. Require segment-specific mapping rather than adding two seconds to every cue.

Score the decision and the evidence separately

Save the reviewer’s expected answer, supporting record IDs and unresolved fields in a separate answer key. Score the model on the decision, its citations and its handling of the deliberate gap. A correct answer citing a different case counts as an evidence error.

Before expanding collection, have a second reviewer run the same checks on the held-out packet. Keep related versions, views and exports of the same case together so they stay on one side of the training and evaluation split.

Check permission and scope

Use independently written scripts and dummy timelines or separately cleared recordings. Exclude client edits, unfinished commercial work, music and protected source files.

Use original or explicitly cleared material and keep third-party music, performer data and unreleased content outside the package.

Describe an authorized archive through Fiund’s existing-data assessment, or describe a new capture capability through the collection brief. The contributor agreement covers the terms for participation.

Sources and further reading

Resources for this workflow

How the examples work

Company and program reviews

Use these reviews to compare sourcing, capture or evaluation approaches. Confirm the applicable program, accepted sources and terms with the provider.

All company reviews

Related industry examples

← Information and publishing