SDH guidelines by platform

SDH is a subtitle track in the same language as the audio. It carries the dialogue and also the sounds a viewer cannot hear: who is speaking, sound effects, and music. This guide covers how to write those parts. For the definitions and the legal differences between subtitles, closed captions, and SDH, read Subtitles vs closed captions.

No single SDH standard exists. Each platform, broadcaster, and regulator publishes its own rules, and they disagree on small things such as brackets versus parentheses, or the music symbol. This guide compares these documents:

• Netflix: English (USA) Timed Text Style Guide, section I for SDH. Last changed 2025.12.19.
• BBC: Subtitle Guidelines, version 1.2.5, March 2026. The BBC calls its closed subtitles "closed captions" too.
• DCMP: the Captioning Key from the US Described and Captioned Media Program.
• CAB: Closed Captioning Standards and Protocol for Canadian English Language Television, 2012 edition.
• FCC: the US caption quality rules in 47 CFR 79.1. This is law, not a style guide.
• Ofcom (UK) and Prime Video: they publish less. See the notes under the tables.

Always ask the buyer which guide applies. When the buyer names none, pick one and use it for the whole file.

How each guide writes the non-speech parts of an SDH file.

And the limits that each guide sets. A dash means that the guide gives no number.

FCC. The US rules set four tests for caption quality: accuracy, synchronicity, completeness, and placement. They require captions to give the non-speech information a viewer cannot see, such as speaker identity, music, sound effects, and audience reaction. They set no character or speed limits.

Ofcom. The 2024 Ofcom guidelines for UK TV and on-demand services no longer set a words-per-minute limit. They ask for subtitles that follow the audio closely, with light editing when speech is very fast. They ask providers to identify speakers clearly (colour, punctuation, or position are the named options) and to move subtitles off the speaker's mouth and other vital information.

Prime Video. Its public subtitling page sets 42 characters per line, 2 lines, and 17 characters per second for adult programs (13 for children's). That page is for subtitles in general. It gives no SDH style rules.

Every guide has the same first rule: name a speaker only when the viewer cannot tell who is talking. The guides differ in how they show the name.

• Netflix puts the name in lowercase brackets, on the same line as the dialogue when possible. For an unnamed character it uses [man], [woman], [male voice], or [female voice], and adds a number when the same label repeats in a scene: [man 1]. Two speakers in one subtitle get a hyphen with no space, with one speaker per line.
• BBC uses colour as its preferred method. White, yellow, cyan, and green are the allowed colours. The narrator is yellow by convention. A capital-letter label is a fallback. Dashes and horizontal placement are legacy methods that new content should not use unless the buyer asks.
• DCMP first uses caption placement under the speaker. If placement cannot work, it puts the name in parentheses on its own line. It says not to name a speaker before the audio or an on-screen graphic introduces them.
• CAB uses a name with a colon, in mixed case, on its own line. It asks you to give the gender in a speaker ID whenever possible.

11200:00:01,000 --> 00:00:03,0003-[Mara] Did you lock the gate?4-Twice.562700:00:03,400 --> 00:00:05,4008[woman 1] Who is at the door?91031100:00:05,800 --> 00:00:07,80012(Mara)13Someone is coming up the path.141541600:00:08,200 --> 00:00:10,20017Mara: Stay behind me.

Subtitle 1 and 2 follow the Netflix pattern. Subtitle 3 follows the DCMP pattern. Subtitle 4 follows the CAB pattern. The BBC pattern needs a text colour, and SRT cannot store it reliably. See Subtitle formats explained for the formats that can.

All guides ask for the sounds that matter to the story. None asks for every sound. The BBC says to skip a sound when the picture or the dialogue already shows it. Netflix says the same for sounds that the visuals imply.

Music has two cases. For music with no words, Netflix and DCMP use a label with a mood or a genre, such as a tense electronic track. The BBC labels known incidental music with MUSIC: and the title. It labels unknown mood music only when it matters to the plot. Netflix also asks you to subtitle silence when it matters, for example when music stops in the middle of a scene.

11200:00:12,000 --> 00:00:14,0003[gate latch clicking]452600:00:12,000 --> 00:00:14,0007LATCH CLICKS8931000:00:12,000 --> 00:00:14,00011(Latch clicking)121341400:00:20,000 --> 00:00:23,50015The ferry leaves at…16[horn blares]171851900:00:23,600 --> 00:00:26,00020…six, so we must hurry.

Subtitles 1 to 3 are one sound in the Netflix, BBC, and CAB styles. Subtitles 4 and 5 show a sound that interrupts a sentence. Netflix ends the first subtitle and starts the next with an ellipsis.

Voice descriptions

DCMP asks for the speaker's emotion when the viewer cannot see it, for example [angrily] or [whispering]. Netflix asks for descriptions of voice, speed of speech, and volume. Netflix also warns against [stutters] unless the speaker has a stutter. Write the hesitation in the text instead.

Every guide in this comparison asks you to subtitle audible lyrics word for word. The BBC says to leave a song unsubtitled when the picture matters more or when the lyrics mix with speech. Netflix subtitles lyrics that do not clash with dialogue. In adult programs, Netflix and Prime Video normally leave theme songs out of subtitles, with an exception for SDH.

Reading speed rules often do not apply to lyrics. DCMP lists sung lyrics among the content that it captions word for word, whatever the rate. The BBC says to time a lyric subtitle to the singing, so it can be much faster or slower than speech.

11200:01:10,000 --> 00:01:13,0003<i>♪ The lamp is low4And the road is long ♪</i>562700:01:13,200 --> 00:01:16,0008<i>♪ Carry me home9Before the song is gone ♪♪</i>

Subtitle 1 follows the Netflix pattern. Subtitle 2 adds the double note at the end that DCMP uses for the last line of a song. Netflix does not use the double note.

Here the guides split into two groups. Netflix and DCMP use italics. The BBC uses quote marks, arrows, and placement.

• Netflix italicizes narration, a visible character's unspoken thoughts, and speech heard through a phone, TV, or speaker when the speaker is not in the scene. It does not italicize a person who is only off camera, behind a door, or out of shot. It also asks you not to switch between italic and roman for the same speaker during a scene.
• DCMP italicizes off-screen dialogue, the narrator when several speakers are present, and off-screen sound effects. If one person speaks in the whole program, DCMP uses roman type.
• BBC uses a single quote at the start of each subtitle for a voice-over, only when the viewer could confuse it with an on-screen voice. For an out-of-vision phone speaker who shares a text colour with the other speaker, it uses single quotes around that person's lines. When the source of an off-screen voice is not clear, it puts an arrow (< or >) toward the speaker's side.
• CAB italicizes disembodied voices and gives them a speaker ID or a description.

11200:02:00,000 --> 00:02:02,5003Hello? Is this the studio?452600:02:02,800 --> 00:02:05,0007[Ines] <i>Yes. We open at nine.</i>

The speaker ID stays in roman type. The italic tags cover only the speech.

• Netflix uses [in Spanish] when the dialogue is translated in the file, and [speaking Spanish] when the viewer is not meant to understand it. It forbids [speaking foreign language]. You must name the language. Mark each language change, but only in the first subtitle after the change. Unfamiliar foreign words inside an English line get italics and no label.
• DCMP says to caption the real foreign words when you can. If you cannot, use a description such as [speaking French]. It says never to translate the words into English.
• BBC covers speech that has translation subtitles in the picture. It puts a language label in capitals, with a colon, above the first of those subtitles. One label is enough for a long run in one language. If the language does not matter, it accepts TRANSLATION:.

Netflix and DCMP differ on one point. Netflix has a pattern for foreign speech that the file translates. DCMP says never to translate it into English. Ask the buyer which one they want.

11200:03:00,000 --> 00:03:02,5003[in Portuguese]4Where did you leave the keys?562700:03:02,800 --> 00:03:05,0008On the table, as always.91031100:03:05,400 --> 00:03:07,40012[speaking Korean]

SDH files carry more text than subtitles, so reading speed is often the hardest limit to meet. The guides give three different kinds of limits.

• Characters per second. Netflix allows up to 20 for adult programs and 17 for children's programs. Prime Video's general subtitling page asks for 17 and 13, and allows 22 as a last resort.
• Words per minute. The BBC recommends 160 to 180 words per minute, which is about 0.3 seconds on screen for each word. DCMP uses 130, 140, and 160 for lower, middle, and upper level educational media. It counts speaker IDs and sound labels as words.
• Time per block of text. The CAB manual sets a minimum of 1.5 seconds for a caption. Its maximum presentation rate is three seconds for each 32 characters of a pop-on caption. DCMP sets a minimum of 40 frames and a maximum of 6 seconds per caption.

Line length also differs. Netflix and Prime Video allow 42 characters per line. The BBC allows 37 for broadcast, because Teletext leaves room for that many. Online, the BBC limits the width of the line and not the character count. The CAB manual limits real-time captions to 32 characters per line, and counts text in units of 32 characters for its off-line timing rules. All of these guides say two lines is the normal maximum.

When text does not fit, the guides tell you to cut words and not to rush the viewer. Netflix says to reduce and condense the text and not to paraphrase. The FCC rule allows paraphrase only when time requires it.

To check your own file, use the CPS calculator for speed and the SRT validator for line length.

The default is the bottom of the screen. The guides then say when to move the subtitle.

• Netflix centres every subtitle and places it at the top or the bottom. It must not overlap on-screen text. If text sits at both the top and the bottom, place the subtitle where it is easier to read.
• BBC uses vertical moves to keep names, captions, and graphics clear. It uses horizontal moves for off-screen direction. It asks you to put important picture information ahead of speaker identification.
• DCMP uses the bottom two lines. It moves the caption to the top when the bottom would cover names, faces, mouths, or key graphics. For an off-screen speaker with a known position, it places the caption at the far left or right.
• CAB says to avoid graphics, eyes, and lips. Move all captions in a scene together, and not one at a time. Pop-on captions can go anywhere on screen, centred or left aligned, never right aligned.
• FCC requires captions that do not block faces, featured text, or graphics, and do not run off the edge of the screen.

Position is a format question as much as a style question. SRT has no position data. WebVTT, TTML, and SCC can store it, and a platform may ignore it. Prime Video's timed text page says that it supports positioning only for some formats. Check the delivery specification before you rely on placement.

A few habits save time on any SDH job. Ask for the guide in writing. Write labels and music symbols the same way in the whole file. Keep a list of speaker names and song titles for the project.

Sometimes a buyer wants both an SDH file and a plain subtitle file from the same source. You can remove the labels, the music symbols, and the speaker names from the SDH file and then check the result by hand.