SDH subtitles: choose and check the right track

What SDH tells you—and what to check

SDH stands for subtitles for the deaf and hard of hearing. An SDH track is intended to convey dialogue plus the relevant sound and speaker information a viewer needs without hearing the soundtrack. A dialogue translation may leave out those details because it assumes the viewer can hear them. Before delivering a file, check its actual coverage against the exact audio version it accompanies.

The label is a useful starting point, not evidence that a file is complete. W3C notes that caption and subtitle terminology varies by region, and that translated subtitles can also include non-speech information. Inspect the content instead of assuming every file called subtitles lacks sound descriptions, or every file called SDH has been reviewed. This guide is for recordings and subtitle files you own or are authorized to adapt and distribute.

Name the soundtrack before choosing the text

Record four things separately: the final video edit, the audio version, the text language and the kind of coverage requested. English text could be a translation of French original dialogue, a transcription of an English dub, or captions prepared for either soundtrack. Matching the language name alone does not establish that the words match what is playing.

Netflix’s English timed-text guide explicitly bases SDH for English-dubbed content on the dubbing script or dubbed audio. That is a Netflix delivery requirement, but it illustrates a useful check for any recipient: ask which audio version the supplied track was reviewed against. A translated subtitle and a dub can express the same meaning with different words, pauses and speaker entrances. Do not describe a mismatch as a timing problem until you have checked that pairing.

Also confirm whether the requested file covers the whole programme or selected translated moments. A forced-narrative subset cannot stand in for full captions. File type and display are separate decisions: an SRT or WebVTT can carry caption text, while the receiving player determines how it is selected and displayed. Renaming a file with SDH does not add missing information or configure the player.

Compare the files at moments that expose a mismatch

Consider an original fictional archive film, The Last Reel. Its French original has an English dialogue translation that reads “Leave the reel here.” The English dub instead says “Put it beside the projector.” Those renderings are invented to show the problem, not taken from a film or produced in an app test. If the recipient wants captions for the English dub, the dub is the audio reference; borrowing the translation without review would put different words on screen.

Later, an already introduced character named Noor answers from another room. The picture stays on the listener, so her identity may need to be conveyed in the captions. Then the projector stops while nobody speaks, and both characters look up. A file containing only spoken sentences can miss the reason for their reaction. These are three different failures: wrong audio wording, unclear attribution and absent meaningful sound.

Use such moments to identify problems in a candidate track quickly. Listen with the proposed file enabled and record the observed discrepancy. A successful spot check is not approval: continue through the entire final recording to find omissions elsewhere. Do not guess a speaker’s identity or a sound source merely to fill the audit sheet.

Fictional track comparison
Video: Last Reel, cut 03
Audio: English dub v2
Text: English candidate A

00:18  Wording differs from dub
00:36  Offscreen reply unclear
00:52  Projector stop missing

Decision: correction required

Keep a coverage sheet outside the subtitle file

Make a small Markdown review note with the event time, what the selected soundtrack supplies, what the current cue conveys and the action needed. Include dialogue, relevant speaker changes, meaningful sounds and music. For each unresolved event, name the question rather than writing a confident caption. If the soundtrack is unclear, ask the producer for the appropriate reference or clarification.

The sheet should distinguish content gaps from technical problems. “No text for the projector stopping” needs an editorial decision and possibly a new cue. “The existing stop caption appears before the sound” needs a timing correction. “The correct file is available but the player selects another English track” needs a destination configuration check. Treating all three as transcription errors can lead you to regenerate text that was already correct.

W3C defines captions around the speech and non-speech information needed to understand the content. DCMP’s sound and music guidance helps with deciding how to describe an event once identified. Use the separate sound-effects guide below for that writing work; this sheet is the acceptance record that makes omissions and mismatched versions visible.

Repair the working copy without losing reviewed material

Keep the supplied file and approved versions unchanged while you prepare corrections. Polyglot can import SRT or WebVTT, review and edit timed text, translate it, and export a separate file on an Apple Silicon Mac. Processing stays local. These capabilities help with the text workflow; they do not establish that a supplied track has complete SDH coverage.

Check existing cues against playback where the matching media is available. Polyglot does not promise automatic speaker identification or a complete inventory of sound effects and music. If a meaningful event has no cue, use a caption editor that supports adding and timing that interval. Do not append a description to unrelated dialogue or stretch a nearby cue just to avoid authoring the missing event.

Keep audio versions as separate review jobs. Translating an approved caption file can provide a draft for viewers of the original soundtrack in another language. It does not automatically make that draft a transcription of a separately recorded dub. When a correction affects translated tracks, review those versions explicitly and keep their previous approved exports until replacements are accepted.

Check the selected track at the destination

Upload or package the reviewed file through the recipient’s supported workflow, then select the intended audio and text together. Exporting from Polyglot does not perform that publishing step. Confirm the displayed track’s language and coverage label, and play through the moments recorded in the sheet. Check the first and last cues as well as the repaired intervals.

A player preference cannot create a missing track. Verify what the viewer can actually choose for this particular video. If several tracks share a language, confirm which entry contains the reviewed full captions and use the destination’s supported labeling options to distinguish it. Do not assume a descriptive local filename will become the viewer-facing label. Record local preview results separately from acceptance on the receiving service.

Watch once with the reference audio to verify accuracy and synchronization, then without sound to look for missing connections. Can the viewer follow the offscreen reply and understand why the characters react? Read at normal speed rather than pausing each cue. Check that extra labels do not crowd dialogue or hide important picture details. The muted pass helps reveal gaps but cannot prove that the text faithfully represents the audio.

Record approval for one specific pairing

Finish the sheet with the reviewed video revision, audio version, text language, exported filename, destination and actual review result. If the destination has not been checked, leave that part pending. Hold delivery when important speech remains unresolved, a meaningful sound is missing, or nobody can establish which soundtrack the file accompanies. A file that imports successfully has passed a different check from one that conveys the programme accurately.

SDH also does not describe every important visual detail for a viewer who cannot see the picture. Visual description and descriptive transcripts serve a separate need. Keep that boundary clear when describing the deliverable: an accurate, usable caption track is valuable, but neither its name nor this checklist certifies the accessibility of the whole video.

Manual acceptance record
Video revision: __________
Audio version: ___________
Text language: ___________
Export filename: _________
Full coverage pass: ______
Open issues: _____________
Player/platform: _________
Destination result: ______
Reviewer and date: _______

Keep exploring.