Subtitles & Captions That Work: Specs, Rates, Accessibility

Most localized videos fail their audiences not because the translation is wrong, but because the subtitles are unreadable. Characters flash by too quickly, lines break mid-phrase, captions lack speaker identification, and formatting collapses when uploaded to a platform that silently strips your carefully authored styles. The result is a technically "localized" asset that frustrates viewers and undermines brand credibility.

This guide is a production-ready reference for building subtitles and captions that actually work, across formats, platforms, languages, and accessibility requirements. It covers the foundational distinctions between subtitle types, the timing and segmentation rules that govern readability, the accessibility mandates you must meet, platform-specific delivery considerations, and the quality checks that separate professional output from amateur work. If your team is scaling video localization across markets, explore how Ollang can streamline your subtitle and caption workflows.

Closed vs Open Captions: When Each Format Applies

The distinction between closed and open captions is fundamentally about viewer control and technical delivery.

Closed captions are delivered as a separate data layer, a sidecar file or embedded metadata track, that viewers can toggle on or off. They are the standard for broadcast television, streaming platforms, and most web video players. Closed captions preserve the viewer's choice and allow a single video file to serve multiple caption tracks in different languages.

Open captions (also called burned-in captions) are rendered directly into the video frames during encoding. They cannot be turned off. Open captions are essential when the delivery channel does not support sidecar tracks, the most common scenario being social media feeds where autoplay is muted and captions must be visible by default to capture attention.

Choosing between them

The decision is rarely about preference. It is driven by the delivery channel, the audience context, and accessibility obligations:

• Broadcast and OTT platforms almost universally require closed captions, often in specific formats mandated by the platform's delivery spec.
• Social media (Instagram Reels, TikTok, LinkedIn native video) either lack sidecar caption support or render it inconsistently, making burned-in captions the practical choice.
• Corporate training and LMS platforms typically support closed captions via WebVTT or SRT, and toggling is expected for multilingual learner populations.
• Live events and conference recordings may use open captions for archival versions to guarantee visibility regardless of player.

A common production pattern is to author captions as sidecar files first, then burn them in as a secondary rendering step for social cuts. This preserves editability and avoids maintaining two separate authoring pipelines.

Subtitles vs SDH: Terminology That Matters

The terms "subtitles" and "captions" are often used interchangeably, but they describe different things, and the distinction has real production implications.

Subtitles assume the viewer can hear the audio. They translate or transcribe dialogue only. They are the standard for interlingual use, a Japanese viewer watching an English-language film with Japanese subtitles.

Captions (and specifically SDH, Subtitles for the Deaf and Hard of Hearing) assume the viewer may not be able to hear the audio at all. SDH includes everything subtitles include, plus non-speech audio cues (a door slamming, music playing, a phone ringing), speaker identification, and descriptions of meaningful sound effects.

What SDH adds to the production scope

Producing SDH requires more than translating dialogue. Your captioning workflow must account for:

• Speaker identification: Labels like [NARRATOR], [DR. CHEN], or positional cues when multiple speakers overlap.
• Non-speech sound cues: Descriptions enclosed in brackets, [suspenseful music], [glass shattering], [crowd cheering], placed in sync with the audio event.
• Music identification: Song titles and lyrics when the music carries narrative meaning, versus a generic [music playing] tag when it does not.
• Tone and manner indicators: Cues like [whispering], [sarcastically], or [speaking in French] when the manner of speech is essential to comprehension.

SDH is required for accessibility-regulated content and is the expected standard for any content intended for broad public distribution.

Key Caption and Subtitle Formats

SRT: Simple, Universal, Limited

SubRip Text (SRT) is the most widely supported subtitle format. Its structure is minimal: a sequential index, a timecode range, and plain text.

1
00:00:01,500 --> 00:00:04,200
Welcome to the quarterly update.

2
00:00:04,800 --> 00:00:07,300
Let's start with the EMEA results.

SRT supports basic formatting (bold, italic, underline) through HTML-like tags, but rendering is inconsistent across players. It does not support positioning, alignment, font specification, or color, which means it cannot reliably handle speaker differentiation through visual styling. SRT uses comma-separated milliseconds in its timecodes, a detail that causes silent failures if a decimal point is substituted.

SRT is the right choice for simple subtitle delivery to platforms like YouTube, Vimeo, and most CMS-embedded players. It is not sufficient for broadcast, accessibility-regulated content, or any use case requiring positional control.

WebVTT: The Web Standard

Web Video Text Tracks (WebVTT) is the W3C standard for web-based video captioning. It is syntactically similar to SRT but adds critical capabilities:

• CSS-based styling: Font, color, size, and background can be controlled via ::cue pseudo-elements in the page's stylesheet.
• Positioning and alignment: Captions can be placed in specific regions of the video frame, essential for avoiding occlusion of on-screen text or graphics.
• Voice tags: <v Speaker Name> allows semantic speaker identification that assistive technologies can parse.
• Header metadata: The WEBVTT header can include descriptive metadata.

WEBVTT

00:00:01.500 --> 00:00:04.200 position:10% align:start
<v Narrator>Welcome to the quarterly update.

00:00:04.800 --> 00:00:07.300
<v CFO>Let's start with the EMEA results.

Note that WebVTT uses period-separated milliseconds, not commas. This is the most common conversion error when migrating from SRT.

WebVTT is the required format for HTML5 <track> elements and is supported by all modern browsers. It is the default choice for web-delivered content where accessibility compliance matters.

IMSC and TTML: Broadcast and Streaming Grade

IMSC (Internet Media Subtitles and Captions) is a constrained profile of TTML (Timed Text Markup Language), the W3C's XML-based subtitle standard. IMSC is the format required by many broadcast regulators and streaming platforms for professional delivery.

IMSC/TTML provides full control over:

• Font family, size, color, background, and opacity
• Precise region-based positioning
• Ruby annotations for CJK languages
• Bi-directional text for Arabic and Hebrew
• Multiple simultaneous caption regions

The tradeoff is complexity. IMSC files are verbose XML documents that require specialized authoring tools and careful validation. They are not hand-editable in a text editor the way SRT files are.

For enterprise localization at scale, the practical approach is to author in a rich format (IMSC or a proprietary tool format) and export to SRT or WebVTT for platforms that require simpler files. When you manage many language versions and output formats, authoring centrally and using a localization platform like Ollang reduces manual conversion and versioning work.

Rates and Pricing: What You Actually Pay For

Rates for subtitles and captions vary across providers, languages, and deliverables. To scope budgets accurately, align on the service bundle and the unit of measure up front.

• Common billing units:
• Per video minute for end-to-end subtitling (transcription, translation, timing, QC)
• Per word for translation-only components
• Per subtitle event or per line in some regions
• Hourly for engineering, format conversions, or complex retiming
• Scope factors that drive cost:
• SDH vs. dialogue-only subtitles (speaker IDs and non-speech cues add effort)
• Audio quality and the need for transcription cleanup
• Domain complexity and required terminology research
• On-screen text (OST) that needs translation and compositing
• Platform deliverables (SRT, WebVTT, IMSC/TTML, SCC) and per-aspect-ratio burn-ins
• Accessibility compliance reviews and second-pass linguistic QA
• Turnaround speed, live vs. prerecorded, and change management after picture lock
• Workflow components that should be explicit in a quote:
• Spotting/timing and segmentation to CPS targets
• Translation with condensation for reading speed
• SDH authoring (speaker IDs and sound cues)
• QC against style guide and platform specs
• Format exports and validation, plus burned-in renders where needed
• Red flags in low quotes:
• No CPS enforcement or segmentation standards
• Machine captions or raw MT with minimal post-editing
• Omitted SDH cues or inconsistent speaker identification
• Single-output pricing when you require multi-format delivery

For predictable budgeting, define your style guide and deliverable list first, then request a bundled per-minute quote for each language. Include room for editorial changes after initial delivery; small timeline edits ripple through timecodes and re-exports.

Timing Rules: Duration, Gaps, and Frame Accuracy

Subtitle timing is where readability is won or lost. Even a perfect translation becomes useless if it appears too briefly, lingers too long, or collides with adjacent cues.

Minimum and maximum display durations

• Minimum duration: A subtitle should remain on screen for at least one second, even for very short utterances. Anything below 0.7 seconds creates a flash that most viewers cannot process.
• Maximum duration: A single subtitle should not exceed seven seconds on screen. Beyond this, viewers lose track of whether the text has changed, and the subtitle feels "stuck." For content with long pauses in dialogue, the subtitle should disappear and reappear rather than persist.

In/out gaps between consecutive subtitles

A minimum gap of two frames (approximately 80ms at 25fps) between consecutive subtitles is the widely accepted standard. This gap gives the viewer's eye a perceptible "blink" that signals new text has appeared. Without it, viewers may not realize the subtitle has changed, especially when the new text is similar in length and position to the previous one.

Some style guides specify gaps as high as 120ms or even 200ms. The right value depends on the frame rate of the source video and the density of the dialogue.

Frame rate and timecode alignment

Subtitles must be authored to the video's native frame rate. A subtitle timed to 25fps will drift when played against 23.976fps content, by roughly four seconds per hour. This is one of the most common causes of "subtitles out of sync" complaints and is entirely preventable with proper source metadata.

For drop-frame timecode (common in NTSC-origin content at 29.97fps), subtitle tools must account for the timecode numbering scheme or risk cumulative drift.

Segmentation and Line-Break Standards

How you break text across lines and across subtitle events has a measurable impact on reading comprehension. Poor segmentation forces the viewer to mentally reassemble phrases, which competes with the visual content for cognitive attention.

Core segmentation principles

• Break at syntactic boundaries. Keep noun phrases, verb phrases, and prepositional phrases intact. Never break between an article and its noun, between a preposition and its object, or between an auxiliary verb and its main verb.
• Two lines maximum per subtitle event. Three-line subtitles are acceptable only in specific broadcast contexts (teletext-era legacy systems) and should be avoided for digital delivery.
• Top-line / bottom-line balance. When a subtitle spans two lines, the bottom line should be equal to or longer than the top line. This creates a visually stable pyramid shape and avoids the top line occluding more of the image than necessary.

Line length limits

Maximum characters per line vary by language and delivery context. A common baseline is 42 characters per line for Latin-script languages, but this is not universal:

These limits exist because subtitle text must remain legible at the video's display resolution. Exceeding them results in text that is either truncated, scaled to an unreadably small size, or wrapped unpredictably by the player.

Handling text expansion

Translation from English into languages like German, French, or Finnish routinely produces text that is 20-35% longer than the source. This expansion must be absorbed through a combination of:

• Condensing the translation (conveying meaning with fewer words, not omitting meaning)
• Splitting a single source subtitle into two target-language events
• Adjusting timing to allow slightly longer display

Condensation is a translation skill, not an editing afterthought. Subtitle translators must be briefed on character limits and reading-speed targets before they begin work, not asked to trim after the fact.

Reading Speed Limits by Content and Audience

Reading speed, measured in characters per second (CPS), is the single most important readability metric for subtitles. A subtitle that exceeds the viewer's comfortable reading speed forces a choice between reading the text and watching the video. Neither outcome is acceptable.

Industry benchmarks

• Adult entertainment (drama, documentary, corporate): 15-17 CPS is the widely accepted comfort zone. Netflix's Timed Text Style Guide specifies a maximum adult reading speed of 17 CPS for most languages.
• Children's content (ages 6-12): 13 CPS maximum. Younger audiences read more slowly and need more time to process both text and image.
• Content for language learners: 12-14 CPS, with simpler vocabulary and shorter subtitle events.
• Fast-paced content (news tickers, sports commentary): Reading speed limits may be relaxed to 20 CPS, but comprehension drops significantly.

CPS versus words per minute

Some legacy style guides use words per minute (WPM) as the reading-speed metric. CPS is more precise because word length varies dramatically across languages. A "word" in German averages significantly more characters than a "word" in English, making WPM comparisons across languages unreliable. CPS normalizes this.

Practical implications

When dialogue is too dense to subtitle within CPS limits, the subtitle translator must condense. This is not optional, it is a core competency of professional subtitle localization. The alternative is subtitles that viewers cannot read, which defeats the purpose of the localization entirely.

If your team is navigating these tradeoffs across dozens of language pairs and content types, see how Ollang's localization platform handles reading-speed compliance at scale.

Accessibility Requirements: WCAG, FCC, and Ofcom

Subtitle and caption accessibility is not a best practice, it is a legal obligation in many jurisdictions, and a practical necessity for reaching the broadest possible audience.

WCAG 2.1 and 2.2

The Web Content Accessibility Guidelines, published by the W3C, establish the global baseline for digital accessibility. For video content:

• WCAG 1.2.2 (Level A): Captions must be provided for all prerecorded audio content in synchronized media. This is the minimum conformance level.
• WCAG 1.2.4 (Level AA): Captions must be provided for all live audio content in synchronized media.
• WCAG 1.4.1: Color must not be the sole means of conveying information, relevant when using color-coded speaker identification in captions.

Level AA conformance is the standard referenced by most government procurement requirements and corporate accessibility policies.

FCC requirements (United States)

The FCC mandates closed captions for video programming shown on television and, through the 21st Century Communications and Video Accessibility Act, extends those requirements to video content that was previously shown on TV and is now distributed online. The FCC evaluates caption quality across four dimensions: accuracy, synchronicity, completeness, and placement.

Ofcom requirements (United Kingdom)

Ofcom requires licensed UK broadcasters to caption a specified proportion of their programming, with targets that increase over time. Ofcom's Code on Television Access Services sets detailed requirements for caption accuracy and presentation.

Beyond compliance

Accessibility-driven captioning also serves viewers watching in noisy environments, non-native speakers using captions as a comprehension aid, and search engines indexing caption text for discoverability. Treating accessibility as a compliance checkbox misses its commercial value.

Speaker IDs, Non-Speech Cues, and Code-Switching

Speaker identification

When multiple speakers appear in a scene, viewers who cannot hear the audio need explicit cues about who is speaking. The standard approaches are:

• Bracketed labels before the dialogue: [MARIA] I disagree with that assessment.
• Color coding (closed captions only): Each speaker is assigned a consistent color throughout the program. This requires a format that supports color (WebVTT or IMSC) and must not be the sole identification method per WCAG 1.4.1.
• Positioning: Placing the caption near the speaker on screen, which requires positional control (WebVTT regions or IMSC).

Consistency is critical. Once a speaker is identified by a label or color, that assignment must persist throughout the entire program.

Non-speech audio cues

Sound effects and ambient audio that carry narrative meaning must be described in SDH captions. The conventions are well established:

• Use square brackets: [phone ringing], [thunder rumbling]
• Place the cue in sync with the audio event, not before or after
• Use lowercase and present participle for ongoing sounds: [dogs barking in distance]
• Use past tense or simple description for discrete events: [gunshot], [door slams]
• Describe music by genre, mood, or title when relevant: [slow jazz music], [plays "Clair de Lune"]

Dialects and code-switching

Multilingual dialogue and code-switching, where a speaker shifts between languages mid-sentence, present a particular challenge. The standard approach is:

• Subtitle the non-primary language in italics (if the format supports it) to visually distinguish it from the primary language
• Add a language tag when the switch might not be obvious: [speaking in Mandarin]
• For SDH, always identify the language switch explicitly, as the viewer may not be able to distinguish languages by ear

Dialect representation in captions is sensitive. Phonetic spelling or exaggerated dialect markers can be perceived as mocking. The general guidance is to represent dialect through vocabulary and syntax choices rather than altered spelling, unless the source material's creative intent requires it and the content owner has approved the approach.

Platform Delivery: Sidecar vs Burn-In on Social and Streaming

Streaming platforms (YouTube, Vimeo, OTT)

YouTube, Vimeo, and major OTT platforms all support sidecar caption uploads. YouTube accepts SRT and SBV natively and auto-generates captions (with well-documented accuracy limitations). Vimeo supports SRT, WebVTT, DFXP, and SCC. OTT platforms typically require IMSC/TTML or WebVTT, specified in their delivery requirements.

For these platforms, sidecar delivery is the correct approach. It enables multiple language tracks, viewer toggle control, and platform-side rendering optimized for the viewer's device and display size.

Social media (Instagram, TikTok, LinkedIn, X/Twitter)

Social platforms present a fundamentally different problem:

• Autoplay with muted audio is the default viewing context. If captions are not visible immediately, viewers scroll past.
• Sidecar support is inconsistent or absent. Instagram Reels and TikTok do not support uploaded sidecar caption files in the way YouTube does. Some offer auto-captioning tools with variable accuracy.
• Aspect ratios vary. A 9:16 vertical frame has much less horizontal space for caption text than a 16:9 frame, requiring shorter lines or larger relative font sizes.

For social delivery, burned-in captions are the standard. The production workflow should:

1. Author captions as sidecar files (SRT or WebVTT) for editability
2. Render burned-in versions per aspect ratio (16:9, 9:16, 1:1) with font, size, and position optimized for each
3. Ensure burned-in text does not collide with platform UI elements (profile icons, like buttons, progress bars)

Safe zones for burned-in captions

Each platform overlays its own UI elements on the video frame. Burned-in captions must avoid these zones:

Testing burned-in captions on actual devices, not just in the editing timeline, is essential. What looks fine on a desktop preview may be occluded on a mobile screen.

A centralized localization platform can automate per-aspect-ratio rendering and sidecar-file generation to reduce manual versioning; tools such as Ollang are designed to centralize that work in the localization pipeline.

Quality Checks and Acceptance Criteria

The QC checklist

A production-ready subtitle QC pass should verify every item on this list:

• CPS compliance: No subtitle event exceeds the target reading speed. Flag any event above the threshold.
• Minimum duration: No event displays for less than one second.
• Maximum duration: No event exceeds seven seconds.
• Inter-subtitle gap: Minimum two-frame gap between consecutive events. No overlapping timecodes.
• Line count: No event exceeds two lines (unless explicitly permitted by the style guide).
• Line length: No line exceeds the character limit for the target language.
• Orphan lines: No subtitle event contains a single word on the second line. Rebalance or merge.
• Segmentation: No line breaks mid-phrase. Articles stay with their nouns, prepositions with their objects.
• Punctuation: Consistent with the target language's typographic conventions (e.g., French uses non-breaking spaces before colons and semicolons; Spanish uses inverted question marks).
• Timecode alignment: Subtitles sync to the video's native frame rate. No drift over the program duration.
• Speaker IDs (SDH): Consistent labeling throughout. No unidentified speakers in multi-speaker scenes.
• Non-speech cues (SDH): All narratively significant sounds described. No missing cues.
• Encoding: File saved in UTF-8 (or UTF-8 with BOM for legacy systems). No encoding artifacts or garbled characters.

Style guide essentials

Every subtitle localization project should be governed by a style guide that specifies:

• Target reading speed (CPS)
• Maximum characters per line
• Preferred line-break behavior
• Punctuation and capitalization conventions for the target language
• Treatment of numbers, dates, currencies, and units
• Italics usage (foreign language, emphasis, off-screen speech)
• Handling of brand names, product names, and untranslatable terms
• Speaker ID format and color assignments (if applicable)

Without a style guide, consistency across translators, episodes, and languages is impossible.

Acceptance criteria

Define pass/fail criteria before production begins, not during review. Common acceptance thresholds include:

• Zero overlapping subtitle events
• Zero events exceeding CPS limit by more than 10%
• Zero events below minimum duration
• All SDH cues present and correctly formatted
• Subtitle file validates against the target format specification (e.g., WebVTT validator passes with no errors)

When Captions Beat Dubbing: A Decision Framework

Subtitles and dubbing are not interchangeable. The right choice depends on the market, the content type, the audience, and the budget.

Markets with strong subtitle preference

Audiences in the Nordic countries, the Netherlands, Portugal, and much of Southeast Asia have strong cultural preferences for subtitled content. Dubbing in these markets can feel unnatural and may actually reduce engagement. Conversely, audiences in Germany, France, Italy, Spain, and much of Latin America have deep dubbing traditions, and subtitled content may be perceived as lower quality or harder to consume.

Content type considerations

The hybrid approach

Many enterprise localization programs use both. A global product launch video might be dubbed into the five largest markets and subtitled into the next fifteen. Training content might be dubbed for factory-floor audiences and subtitled for office-based teams. The decision is not binary, it is a portfolio strategy.

The key is to make the decision early in production, because dubbing and subtitling impose different constraints on script writing, recording, and post-production timelines. Not sure whether to dub, caption, or mix both? Get a workflow recommendation for your markets.

Frequently Asked Questions

What is the difference between closed captions and subtitles?

Closed captions are a toggleable text layer that includes dialogue, speaker identification, and non-speech audio descriptions, designed for viewers who cannot hear the audio. Subtitles typically include only dialogue translation and assume the viewer can hear the audio. SDH (Subtitles for the Deaf and Hard of Hearing) bridges the two by adding accessibility elements to subtitle-style delivery. The format choice depends on the delivery platform, the regulatory context, and the audience's needs.

What reading speed should I target for subtitles?

For adult audiences watching drama, documentary, or corporate content, target 15-17 characters per second (CPS). For children's content, stay at or below 13 CPS. These are not arbitrary numbers, they reflect the speed at which viewers can comfortably read text while still engaging with the visual content. Exceeding these limits forces viewers to choose between reading and watching, which degrades the experience.

Should I use SRT or WebVTT for my subtitle files?

SRT is the simplest and most widely compatible format, suitable for basic subtitle delivery to platforms like YouTube and Vimeo. WebVTT adds styling, positioning, and semantic speaker tags, making it the better choice for web content where accessibility compliance (WCAG) matters. If you are delivering to broadcast or premium streaming platforms, you will likely need IMSC/TTML. At scale, use a localization platform to author centrally and export to each required delivery format; tools such as Ollang are commonly used to centralize that workflow.

When should I burn captions into the video instead of using a sidecar file?

Burn-in is the right choice when the delivery platform does not reliably support sidecar captions, primarily social media platforms like Instagram Reels, TikTok, and LinkedIn native video, where autoplay is muted and captions must be visible by default. For all other contexts (web, OTT, broadcast, LMS), sidecar files are preferred because they preserve viewer control, support multiple languages, and are easier to update without re-rendering the video.

Get Started with Production-Ready Subtitles and Captions

Building subtitles and captions that meet technical specifications, accessibility requirements, and audience expectations across every target language and platform is a complex production challenge. It requires the right combination of linguistic expertise, format knowledge, timing precision, and platform-specific delivery logic.

Ollang's localization platform handles the full subtitle and caption workflow, from transcription and translation through timing, QC, and multi-format delivery, so your team can scale video localization without sacrificing quality or compliance.

Ready to scale production-ready subtitles and captions across formats, languages, and channels?

Book a Demo