A local history center is preparing a recorded interview about an old textile mill for a public screening. The words need to be accessible, but a familiar automatic-transcription mistake could change a resident’s name or place a warning on the wrong shot. Amazon Transcribe can generate WebVTT and SubRip caption files during a batch transcription job. That saves a mechanical formatting step; it does not authorize publication or prove the words and timing are correct. The film, names and examples here are fictional educational material, not tested audio or measured caption accuracy.
Keep an immutable picture-and-sound reference
Before requesting captions, assign the edited film a stable version and preserve the source audio. Record its running time, frame rate, intended audience, language and existing speaker permissions. A subtitle proof must refer to exactly the video that viewers will see; if an editor later removes a pause or reorders a scene, previously accepted cue times may no longer match. Retain a transcript with clear uncertainty marks for hard-to-hear names instead of filling them from a plausible memory.
Make a small review sheet for names, locations and archival terms with links to approved records. Do not send contact details, private interviews or unlicensed music to an external service just because subtitles would be useful. Separate the public video from consent paperwork and private source recordings. If the source film contains a safety or accessibility announcement, identify it as a mandatory cue for final review rather than trusting the subtitle system to make the announcement prominent.
Choose the batch subtitle formats deliberately
Amazon Transcribe’s subtitle feature can produce WebVTT files with a .vtt extension and SubRip files with a .srt extension alongside the ordinary transcript. Select either or both formats with the Subtitles Formats option of a batch job. The files are delivered to the same configured destination as the transcript. The service creates subtitles only when speech is present; a film with no detectable speech may produce no subtitle file, so the absence of output must not be treated as successful delivery.
Choose the destination bucket and access rights before processing. Retention policies should cover intermediate text, generated files and the final published version. Test the consuming video player with the selected format; support varies by workflow, so do not call one format universally better. This is a source-based procedure, not a claim that we uploaded a video or inspected actual Amazon output.
Do not overlook OutputStartIndex
The AWS Subtitles API documents OutputStartIndex as either 0 or 1. Amazon Transcribe defaults to 0, while many downstream subtitle tools expect the more common index 1. A file can have plausible cue text and timestamps yet be rejected by another tool over its numbering convention. Set this option intentionally, ideally based on the actual editor or player specification, rather than altering only the visible heading of a document.
A hypothetical SRT comparison should show cue 0 in the first generated file, a compatibility warning during import and a reviewed cue 1 sequence for a pipeline that expects it. This illustrates an integration risk, not a tested failure. Cue numbering is distinct from the time code and separate from language or transcript correctness. If a renderer accepts zero-based cues, document that requirement instead of normalizing blindly.
Compare caption times with the spoken track
Use the generated subtitle file as a timing proposal. Scrub through the film while listening, especially at starts, ends, pauses and rapid changes between speakers. A sentence split across cues may be grammatical but difficult to read; a combined cue may remain on screen too long after a speaker stops. AWS explains that its subtitles follow speech and end around natural pauses, but publication still requires editorial review at the actual playback speed and screen size.
Inspect the film’s final export, not only a transcript-only view. Verify that lower-third names and captions do not obscure each other and that an on-screen archival date is not contradicted by the dialogue caption. If a speaker whispers a name, leave a query for the archival reviewer rather than replacing it with an unsupported proper noun. For multilingual excerpts, consult someone competent in both the speech and the intended caption language.
Handle speaker changes and masked words honestly
AWS documents speaker-diarization support in subtitles: when a segment contains multiple speakers, dashes help distinguish them. This is a formatting aid, not verified identity attribution. In an interview with an interviewer and witness, the review team should listen to turns and check whether a dash is attached to the right speech. Do not invent a personal name from an anonymous speaker label or assume that an interviewer obtained every needed license.
Content redaction and vocabulary filters propagate into subtitle text when configured, but neither edits the underlying audio. AWS notes redacted information can appear as PII and filtered words may become asterisks or whitespace. That creates a critical mismatch if viewers still hear a private address. Decide whether the public soundtrack needs a separately approved audio edit, with a new video version and new caption proof, before calling the publication privacy-safe.
Check reading comfort, not just transcript agreement
Review line breaks, speaker cues, punctuation and display duration using the actual exhibition player. A long caption on a fast scene can leave a reader behind even if every word matches the audio. Accessibility review should consider whether background sound, off-screen speakers and meaningful non-speech cues require description under the center’s caption policy. A transcript file does not automatically satisfy every accessibility requirement of a finished film.
Have a reviewer compare selected difficult passages against the raw recording and the approved source notes. Add a second reviewer for proper names, dates and historically contested claims. Preserve corrections as versioned editorial changes; a person should be able to tell what was generated, what was changed and which film export was approved. Do not report improvement rates or caption accuracy without a measured reference set.
Test the import path before screening day
Prepare a short, permission-cleared test clip with two distinct speakers, a pause and a proper noun. Request both supported file formats and examine their syntax, numbering and load behavior in the actual display software. Check the case where the subtitle file is missing, the video has no speech or the player silently ignores a malformed file. A successful API job is not proof that the auditorium screen will show the captions.
Verify the media and caption objects are both accessible to the intended playback system but not exposed to unrelated users in cloud storage. If the import fails, use a manual fallback with reviewed captions rather than showing inaccurate subtitles. Test after changing the video encoder, player or export format. Keep the test synthetic unless the organization has the right to use real recordings for that purpose.
Release a captioned film with a clear audit trail
Publish only the final video version together with the matching reviewed VTT or SRT. Store its film revision, caption revision, reviewer, date and any unresolved uncertainty in the editorial log. If the film is re-cut, invalidate the subtitle sign-off until the times are checked again. If an interview subject’s permission changes, update the content policy and delivery locations; caption approval cannot override consent or copyright.
The figures in this article are original conceptual diagrams, not Amazon console screenshots or results from a live job. Check AWS format and API availability, storage costs, regional service constraints and your legal obligations before deploying. The operational objective is a faithful, readable caption track tied to the public film—not simply two files bearing familiar extensions.
How We Research
For this November 2026 article, we compared named official references: AWS: Creating video subtitles; AWS: Subtitles request parameters; AWS: StartTranscriptionJob API. We checked the documented request shape, output behavior, and important limits; then we developed an original hypothetical nonprofit workflow with source checking and explicit human authorization. This is documentation-based editorial research, not a laboratory test: we did not upload real files, run the vendors, measure accuracy or claim real customer outcomes. The illustrations are educational schematics, not platform screenshots or measured results. Before implementation, verify current API access, regional features, security, costs, consent and publication policies.
Frequently Asked Questions
It can generate one or both formats in a batch transcription job if requested; always inspect the produced files.
Amazon defaults OutputStartIndex to 0. Some downstream applications expect 1, so set the value deliberately.
No. AWS documents that content redaction and vocabulary filtering affect the text output; the audio requires separate review and editing.
No. The formatting helps distinguish turns; identity and consent still need source verification.
No. Verify names, times, player behavior, accessibility and the exact film revision before release.