Check Subtitle Speaker Identification
Or paste subtitle text
Options
The limit is not a number you type in: it is read out of a published document, and the report names that document, quotes the sentence it came from, says whether that sentence is a requirement or guidance, and gives the day we last verified it with a link. Pick the document you are answerable to. They do not agree with each other: Netflix counts characters per second and caps a line at 42 characters, while Ofcom and the BBC count words per minute and, on line length, the BBC caps a line at 37. Three of them state no figures at all — 47 CFR §79.1, WCAG 2.2 SC 1.2.2 and EN 301 549, which requires web pages and non-web documents to satisfy that same criterion. Those three require captions to identify who is speaking and to carry non-speech information, and leave every number to the broadcaster; a page that checks them asks whether the file does it, not whether some number is exceeded. They also do not carry the same weight: a defect that reads as a warning against WCAG reads as an error against EN 301 549, because the standard turns that criterion into a shall. Changing this box changes the limit and the citation together, so a report can never carry one document's number under another document's name. Pages that check something only one of these documents states offer only that one.
Which frame rate the timings are counted in. On the format pages this is what MicroDVD needs, because it stores timing as frame numbers. On the frame rate pages it is the rate the file was timed against — the version the subtitles were made for — and the page's second box takes the rate of the video you are going to use them with. 23.976, 24, 25, 29.97, 30, 50, 59.94 and 60 are the common ones. Getting it wrong shifts the whole timeline, and on the frame rate pages it shifts it by a ratio rather than by a fixed amount.
When two people talk inside one subtitle cue, the cue has to say so, and it has to say who. A dash at the start of a line marks the change of speaker; a name in front of the line names them. This page lists the cues that do the first and skip the second. The file is read in your browser, never uploaded, and is not changed — what comes back is a report you can download as a PDF or as a CSV.
The three documents this page can judge against say the same thing and give no figures at all. FCC 47 CFR §79.1(j)(2)(i) requires captioning to provide nonverbal information that is not observable, naming the identity of speakers as an example; WCAG 2.2 Success Criterion 1.2.2 asks that captions identify who is speaking. EN 301 549 adds nothing of its own — its clause 9.1.2.2 requires web pages, and 10.1.2.2 non-web documents, to satisfy that same criterion, which is the route the European Accessibility Act takes to subtitles — but it is the document a European service is assessed against, so the report names it and quotes its clause number. None of the three says how many turns a cue may hold, what a name looks like, or how long any of it should be on screen. So the report has no Limit column — not a column of dashes, not a zero, but no column, because there is no number to print and printing one would mean inventing it.
That leaves the hard part, which is deciding what counts as a defect. This page reports a cue that holds two or more speaker turns and names none of them, and it will only do so for a file that names speakers somewhere else. A file that never names anybody is reported as not applicable, with the reason stated. The reasoning is in the note on that example below, and it is the most important thing on this page.
Examples
Two speakers, two dashes, no names
Subtitles
1
00:00:01,000 --> 00:00:03,000
JOHN: I told you not to wait up.
2
00:00:04,000 --> 00:00:07,000
- I know what you did.
- Then say it out loud.
3
00:00:08,000 --> 00:00:10,000
MARY: It does not matter now.
What the report finds
Cues with two speakers and no name — 1 found
Severity Cues Start End Measured Clause
Error 2 00:00:04.000 00:00:07.000 2 items 47 CFR §79.1 — Closed captioning of televised video programming — (j)(2)(i) Accuracy: captioning shall provide nonverbal information that is not observable, such as the identity of speakers
Checked against
Standard: FCC caption quality rules (47 CFR §79.1)
Where it is stated: (j)(2)(i) Accuracy: captioning shall provide nonverbal information that is not observable, such as the identity of speakers
Stated as: Requirement
Not checked by this page: (j)(2)(i) Accuracy: captioning shall provide nonverbal information that is not observable, such as the existence of music, sound effects, and audience reactionOnly the middle cue is reported. The other two name their speakers, which is what makes the file checkable at all: a file that names speakers in some cues has a convention, and the middle cue breaks it. Notice the columns: Measured is there and reads 2 items, because two turns were counted; Limit is absent, because 47 CFR §79.1 never states a figure. The severity reads Error rather than Warning, and that is also the document talking — this sentence is a regulation, where the Ofcom, Netflix and BBC limits on other pages are guidance their authors wrote for themselves.
One dashed line is one person talking
Subtitles
1
00:00:01,000 --> 00:00:03,000
MARY: I told you not to wait up.
2
00:00:08,000 --> 00:00:10,000
- And that is the end of it.
What the report finds
Cues with two speakers and no name — Nothing found.
Checked against
Standard: FCC caption quality rules (47 CFR §79.1)
Where it is stated: (j)(2)(i) Accuracy: captioning shall provide nonverbal information that is not observable, such as the identity of speakers
Stated as: Requirement
Not checked by this page: (j)(2)(i) Accuracy: captioning shall provide nonverbal information that is not observable, such as the existence of music, sound effects, and audience reactionThe boundary this page is built around: a dash marks a change of speaker, so one dashed line is one person talking and there is nobody to identify. Two is the smallest number that counts. Written as "more than one turn" in one direction and "at least two" in the other, this fixture would be the one that goes red — and the same off-by-one on this page would either flag every ordinary cue in a file or pass every cue that holds a conversation.
A file that never names anyone
Subtitles
1
00:00:01,000 --> 00:00:03,000
- I know what you did.
2
00:00:04,000 --> 00:00:07,000
- Then say it out loud.
What the report finds
Cues with two speakers and no name — Not applicable. This file never names a speaker, so there is no way to tell whether the ones it leaves unnamed should have been named
Checked against
Standard: ETSI EN 301 549 (EU accessibility requirements for ICT)
Where it is stated: 9.1.2.2 Captions (Prerecorded): Where ICT is a web page, it shall satisfy the WCAG 2.1 Success Criterion 1.2.2 Captions (Prerecorded)
Stated as: Requirement
Not checked by this page: 10.1.2.2 Captions (Prerecorded): Where ICT is a non-web document, it shall satisfy the WCAG 2.1 Success Criterion 1.2.2 Captions (Prerecorded)"Not applicable", not "nothing found", and the difference is the point of the whole page. The sentence being applied requires captioning to supply information that is not observable, and whether the viewer can see who is talking is not something a subtitle file records — so a two-dashed cue in a file that names nobody is not evidence of anything. What a file can show is that it names speakers elsewhere, which is evidence that identity is not always visible in this material. A dashed-through file offers no such evidence, and reporting zero findings would have said the same thing as passing it. This example is checked against EN 301 549, and the last two lines of the report are why that standard is in the list: the clause it applies is quoted with its number, and the line below names the clause it does not apply. Switch the standard to WCAG and the verdict is identical, wording and weight aside.
Frequently asked questions
Which documents can speaker identification be checked against on this page?
Three. FCC 47 CFR §79.1(j)(2)(i) requires captioning to provide nonverbal information that is not observable, and names the identity of speakers as an instance of it; WCAG 2.2 Success Criterion 1.2.2 requires that captions identify who is speaking; EN 301 549 §9.1.2.2 requires web pages, and §10.1.2.2 non-web documents, to satisfy that same criterion, which is how the European Accessibility Act reaches subtitles.
They do not carry the same weight, and the report says so in a row of its own: the FCC rule and the EN 301 549 clause are stated as requirements, where the WCAG sentence this page applies sits in the Understanding document, whose own header describes its explanations as informative and not required to meet WCAG. So the same cue reads as an error against the first two and as a warning against WCAG. Wording aside, all three state a duty rather than a number, which is why the report has no Limit column.
Why do two dashed lines in one cue count as two speakers rather than one?
Because the dash is the convention for a change of speaker: a subtitle line beginning with a dash is understood to be a different person from the line above it. That convention is what makes a two-dashed cue worth reporting — the reader is told that somebody else is talking, and is not told who. Both conventions are recognised here: a line beginning with a dash counts as a turn, and a line beginning with a name and a colon counts as a turn that is named.
Why is a file that never names a speaker reported as not applicable rather than clean?
Because whether a speaker needed naming is not something a subtitle file can answer. The requirement is about information the viewer cannot observe; if the speaker is on screen with their lips moving, there is nothing to supply, and a text file records nothing about what was on screen. A file that names speakers in some cues is evidence that identity is not always visible in that material, which is why the convention exists there and can be held to. Where no such convention exists anywhere in the file, this page says it cannot judge, and states that reason, rather than reporting zero findings — a clean report and a check that could not be made read identically otherwise, and they are not the same answer.
Does 47 CFR §79.1 say how many speakers a single caption may hold?
No. The FCC caption quality rules set out four qualities — accuracy, synchronicity, completeness and placement — and the identity of speakers appears under accuracy. There is no figure anywhere in the section: no number of turns per cue, no character limit, no rate. That is why the report for this page prints what it measured and leaves the limit column out entirely, and it is worth knowing before you go looking for a number to put in a house style document.
Is a music or sound label treated as a speaker name on this page?
No. A line like MUSIC: Slow piano. has the shape of a named turn, and counting it as one would make a perfectly ordinary cue look as though somebody had been identified. Labels of that kind — music, note, sound, lyrics — are kept out of the count on both sides. They are non-speech information, which is a separate requirement in the same sentence and a separate check; this page lists it under what it did not check.
Does checking speaker identification change my subtitle file?
No. The file is read and left alone, and nothing is uploaded. If you want names removed, that is the opposite tool, and it writes a new file. If you want the two speakers in a reported cue split into two cues, that is a timeline edit rather than something this page can decide for you, since only you know where the turn actually falls.
What the check looks for
- The report describes a file, not a service. 47 CFR §79.1 binds television video programming distributors in the United States; WCAG 2.2 is a standard a site or service is assessed against, and EN 301 549 is the harmonised standard the European Accessibility Act points at. None of the three is a statement about a subtitle file on its own, and nothing here concludes that your subtitles comply. What the report states is how many cues measured the way this page measures them, which sentence each finding cites, and which parts of those documents this page did not check at all.
- This check is deliberately the narrow one, and the narrowness is the honest part. Speaker identification is required where the identity is not observable — a fact about the picture, which a subtitle file does not carry. Rather than guess, this page uses the file's own naming convention as the evidence that naming was needed. The consequence is that files which never name speakers are never reported as failing, and that is a real limit of the check rather than a feature of it.
References
- 47 CFR §79.1 — Closed captioning of televised video programming, paragraph (j)(2)(i): captioning shall provide nonverbal information that is not observable, such as the identity of speakers. These are the FCC caption quality rules, quoted from the text as published.
- WCAG 2.2 Success Criterion 1.2.2 (Captions, Prerecorded), Level A. The criterion itself does not mention identifying who is speaking; that sentence sits in the Understanding document for it, whose header states that its explanations are informative and not required to meet WCAG, which is why this page records it as guidance rather than as a requirement.
- EN 301 549 (ETSI) v3.2.1 — clause 9.1.2.2 Captions (Prerecorded): where ICT is a web page, it shall satisfy the WCAG 2.1 Success Criterion 1.2.2; clause 10.1.2.2 says the same for non-web documents. The standard states nothing of its own about captions beyond that incorporation, which is why the sentence this page applies is quoted from WCAG and only the clause number is quoted from here.