Subtitling and dubbing

In short

Subtitling and dubbing are the two main modes of audiovisual translation (AVT): subtitling renders speech as timed written text on screen, dubbing replaces the original dialogue track with a performed recording in another language. Both are constrained translation — the translator works against the clock, the shot and the lip, not just the sentence, so condensation and synchrony matter as much as meaning. They are examined in BA/MA papers on translation and media (modes of AVT, constraints, spotting, lip-sync types, accessibility) and tested in placement rounds for OTT localisation vendors, dubbing studios and media houses. In India the demand is driven by pan-India film releases, OTT catalogues in 8–12 Indian languages, anime and Korean content dubbing, and by legal accessibility duties under the Rights of Persons with Disabilities Act, 2016 and the Ministry of Information and Broadcasting's accessibility standards for feature films in theatres. Typical entry roles are subtitler, QC/quality checker, dialogue adapter, dubbing assistant and language lead.

Explained

What the two modes are, and the family they belong to

• Subtitling: condensed written translation of speech (and relevant sound) displayed in sync with the image, usually at the bottom, 1–2 lines.
• Dubbing (lip-sync revoicing): the original dialogue is removed and re-recorded by actors in the target language, synchronised to lip movement and body action; the music and effects are retained.
• Voice-over: the original voice stays audible and is overlaid by a translated reading, usually starting a second late and ending a second early (documentaries, news, interviews).
• Narration / commentary: a free, rewritten spoken text, not bound to the original utterance.
• Interlingual (across languages) vs intralingual (same language — captions for the deaf, karaoke, language learning) vs intersemiotic (speech → sign language; image → audio description).
• Open subtitles are burnt into the picture and cannot be switched off; closed subtitles/captions are a separate data stream the viewer toggles.
• SDH (subtitles for the deaf and hard-of-hearing): adds speaker identification, sound effects ("[door creaks]"), music cues and tone; CC (closed captions) is the broadcast equivalent, verbatim-leaning.
• Forced narrative (FN) subtitles: translate on-screen text (signboards, SMS, letters) or foreign-language snatches even when the viewer has chosen the same language as the audio.
• Audio description (AD): a spoken track describing visual action for blind viewers — delivered in gaps between dialogue.

Subtitling: the process, step by step

1. Material in: video with burnt-in timecode (BITC), audio stems, a dialogue list or as-broadcast script, a KNP list (key names and phrases), and style guide.
2. Transcription / template creation: a verbatim transcript of the source language, or an English template (a master timed file) used as a pivot for many target languages — common in Indian OTT work, and the reason relay errors spread.
3. Spotting (cueing): fixing in- and out-times of each subtitle to speech rhythm and shot changes. A subtitle should not cross a hard cut; if it must, guides allow it only when the cut is more than about 12 frames from the boundary. Leave a 2-frame (or 2-3 frame) gap between consecutive subtitles.
4. Translation and condensation: render the meaning within the character budget — typically 30–40% of spoken words are dropped (repetitions, fillers, vocatives, redundant deictics) without losing information carried by the image.
5. Line breaking and segmentation: break at syntactic joints — not between an adjective and its noun, an auxiliary and its verb, or a preposition and its object; keep the top line shorter if the sentence allows and avoid orphan words.
6. Review / QC: language check, reading-speed check, sync check, glossary and consistency check, format validation, then simulation (watching the film with subtitles on).
7. Delivery: export in the required format at the required frame rate, with a QC report.
8. Technical parameters (industry norms, vendor guides differ): maximum 42 characters per line (37 in classic European TV practice), maximum 2 lines; minimum duration about 5/6 second (20 frames at 24 fps); maximum about 7 seconds; reading speed about 17 characters per second (CPS) for adults, 13 CPS for children's content; some Indian-language guides allow 20 CPS. Confirm against the client's current style guide — these numbers are contractual, not universal.
9. Conventions: italics for off-screen voices, songs and foreign words; a dialogue dash when two speakers share one subtitle; numerals, units and currency written per style guide (₹, lakh/crore or million as specified); songs subtitled only if plot-relevant and rights permit.
10. Indic-script issue: character counts in Devanagari, Tamil, Malayalam, Bengali and Kannada depend on whether you count code points or grapheme clusters (matras, conjuncts, viramas). The same sentence can pass or fail a 42-character check depending on the tool, so agree the counting rule and use NFC-normalised Unicode.

Subtitle file formats and timecode

• Timecode is HH:MM:SS:FF (frames) or HH:MM:SS,mmm (milliseconds, as in SRT). Frame rates: 24 fps (cinema/DCP), 25 fps (PAL-legacy Indian TV), 23.976 and 29.97 fps (NTSC-derived masters), 30 fps.
• Conforming/retiming is required when the frame rate or cut changes; a wrong frame-rate assumption causes a drift that grows through the film — the single commonest delivery rejection.
• Free and professional tools: Subtitle Edit, Aegisub, Subtitle Workshop, Ooona, EZTitles, SubtitleNEXT, plus web platforms used by vendors.

Dubbing: the process and types of synchrony

1. Detection / time-coded script: the original is broken into loops (short takes), each with timecode, speaker, and notes on lip-visibility.
2. Translation: a literal, complete rendering with notes — often by a translator who is not the adapter.
3. Dialogue adaptation (dubbing-script writing): rewriting each line to fit duration, syllable count, visible labials and the character's register — the craft stage, and the skill employers pay for.
4. Casting and voice test: matching vocal age, timbre and energy; a star's "voice" in a dubbed market often becomes fixed across films.
5. Recording: actors record loop by loop against picture, with beeps or a rythmo band (a scrolling text strip synchronised to the image, the French/Canadian method). ADR is the same technique used to replace dialogue in the original language.
6. Editing and mixing: dubbed dialogue is mixed with the M&E track (music and effects) supplied by the rights-holder; Foley and loop-group walla are added if missing; stems are delivered 5.1 or stereo.
7. QC and certification: sync and lip check, mix check, then — for theatrical release in India — censor certification of that dubbed version.
8. Types of synchrony: phonetic/lip sync (matching visible bilabials /p, b, m/ and open vowels), isochrony (line starts and ends with the mouth; the strictest numeric constraint), kinesic sync (matching gestures, nods, head turns), character sync (voice fits the body on screen), content sync (meaning fits the visible action).
9. Advantages of dubbing: no reading load, works for children and low-literacy audiences, preserves the image, immersive. Limitations: 10–30× the cost of subtitling, loses the original performance, lip mismatch is distracting, slower turnaround.
10. Advantages of subtitling: cheap, fast, keeps the original voices, aids language learning and accessibility. Limitations: heavy condensation loses nuance, covers part of the image, needs literacy and reading speed, poor for action-dense scenes.
11. Dubbing countries vs subtitling countries: historically Italy, Germany, Spain, France dubbed; Scandinavia, Netherlands, Greece subtitled. India does both at scale and simultaneously — a pan-India release ships dubs in 4–6 languages plus subtitles in 10+.

Indian practice, law and accessibility

• Copyright Act, 1957: subtitling and dubbing are exercises of the translation and adaptation rights (s.14) in the underlying literary/dramatic work and in the cinematograph film, so a licence from the rights-holder is essential; the subtitle file or dub script is itself a protected derivative work, usually assigned to the commissioner under a work-for-hire clause. Moral rights (s.57) survive assignment. There is a specific exception permitting conversion of a work into an accessible format for persons with disabilities (s.52(1)(zb)).
• Cinematograph Act, 1952, as amended by the Cinematograph (Amendment) Act, 2023: every dubbed version needs its own CBFC certificate; the 2023 amendment introduced the age-based sub-categories UA 7+, UA 13+, UA 16+, made certificates perpetually valid, and added anti-piracy offences for unauthorised recording and exhibition — relevant because leaked dub masters and camera-rip subtitles are a live piracy route.
• IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021: OTT publishers self-classify content as U / U/A 7+ / U/A 13+ / U/A 16+ / A with content descriptors, and are expected to improve accessibility features for persons with disabilities; the classification travels with every dubbed and subtitled version.
• Rights of Persons with Disabilities Act, 2016 (s.42): access to electronic and audiovisual media must be ensured through audio description, closed captions and sign-language interpretation.
• MIB accessibility standards for public exhibition of feature films (issued 2024): films certified for theatrical release are to carry embedded closed captions and audio description for hearing- and vision-impaired viewers, on a phased timeline. The phasing has been revised more than once — state the duty, and check the current notification for the applicable date and which releases are covered.
• Indian Sign Language interpretation is coordinated through ISLRTC; a growing deliverable on government and public-broadcast content.
• Industry: dubbing hubs are Chennai, Hyderabad, Mumbai, Kochi and Bengaluru; localisation vendors include Prime Focus/DNEG localisation arms, Sound & Vision India, VR Films, Main Frame, Pixelogic and Deluxe-type global houses, plus many studio-level SMEs. Buyers are Netflix, Amazon Prime Video, JioHotstar, ZEE5, SonyLIV, Aha, Sun NXT, Hoichoi, Chaupal and the film producers themselves. Voice artists organise through bodies such as the Association of Voice Artists (AVA), which has pressed for consent and payment norms on AI voice cloning.
• Commercials and tax: work is quoted per runtime minute (subtitling) or per minute plus shifts (dubbing); a recording shift is conventionally 3–4 hours. Services attract GST at 18%; subtitling/dubbing for a foreign client is normally an export of services, zero-rated against an LUT. Clients deduct TDS on professional fees (10% for residents) under the TDS provisions of the Income-tax Act, 2025 (in force from 1 April 2026). GST registration is required once service turnover crosses the threshold (₹20 lakh, ₹10 lakh for special-category states).
• AI in the loop: ASR auto-transcription, machine-translated subtitle drafts, synthetic voices and AI lip-sync are now standard first passes, with human post-editing and QC on top; India's Bhashini mission and AI4Bharat's Indic models feed Indian-language pipelines. Quality, consent and credit for cloned voices remain contested.

Key points

• AVT - audiovisual translation; subtitling, dubbing, voice-over, narration, SDH, audio description.
• Constrained translation - the text must fit time, space, image and lip, so equivalence is negotiated, not literal.
• Spotting/cueing - setting in and out times; respect shot changes and the 2-frame gap.
• CPS - characters per second; about 17 for adults, 13 for children; roughly 170 words per minute at 17 CPS.
• Character limit - 42 characters per line, maximum 2 lines, in most current OTT guides (37 in classic TV practice).
• Six-second rule - two full lines held about six seconds, the older broadcast reading-speed yardstick (~12 CPS).
• Condensation - 30-40% of spoken text is typically dropped; cut fillers and redundancy, never new information.
• Segmentation - break lines at syntactic units; never split a noun phrase or verb group.
• Forced narrative - subtitles for on-screen text and foreign-language dialogue, shown even in the same-language track.
• SDH vs CC - both serve deaf viewers; SDH is a subtitle stream, CC is broadcast-encoded (CEA-608/708) and more verbatim.
• Template/pivot translation - an English master file drives many languages, cheap but propagates errors.
• Isochrony, phonetic sync, kinesic sync - the three synchrony demands on a dub line.
• M&E track - music and effects without dialogue; without it, dubbing requires rebuilding the soundscape.
• Loop / ADR / rythmo band - the unit of recording, the technique, and the scrolling-text aid.
• Frame rates - 24 (cinema), 25 (Indian TV), 23.976/29.97 (NTSC-derived); a wrong assumption drifts through the film.
• Legal spine - Copyright Act, 1957 (translation/adaptation rights, s.57 moral rights, s.52(1)(zb) accessible formats); Cinematograph Act, 1952 as amended in 2023 (separate certificate per dubbed version, UA 7+/13+/16+); IT Rules, 2021 self-classification; RPwD Act, 2016 s.42.
• Cost ratio - dubbing costs roughly 10-30 times subtitling per minute, which is why subtitles ship first.

Formulas and examples

Example 1 — does the subtitle fit? A Malayalam line is translated into English as 78 characters (with spaces) and the shot allows 3.4 s.

• CPS = 78 / 3.4 = 22.9 CPS — above the 17 CPS limit, so it fails.
• Allowed text: C_max = 17 × 3.4 = 57.8 → 57 characters; cut 78 − 57 = 21 characters (27% condensation).
• Or, if the shot permits, extend to t_min = 78 / 17 = 4.6 s.
• At 22.9 CPS the viewer is being asked to read ≈ 229 WPM — comfortable silent reading is ~200 WPM, and the viewer must also watch the image.

Example 2 — timecode and duration at 25 fps. In 01:12:33:10, out 01:12:36:22.

• F_in = ((1×3600) + (12×60) + 33) × 25 + 10 = (3600 + 720 + 33) × 25 + 10 = 4353 × 25 + 10 = 108,835 frames.
• F_out = (3600 + 720 + 36) × 25 + 22 = 4356 × 25 + 22 = 108,922 frames.
• Duration = (108,922 − 108,835) / 25 = 87 / 25 = 3.48 s.
• Subtitle of two lines, 40 + 32 = 72 characters → CPS = 72 / 3.48 = 20.7 CPS, over limit; C_max = 17 × 3.48 = 59.2 → 59 characters.

Example 3 — frame-rate drift. A file timed against a 24 fps master is played back as 25 fps.

• Drift = 4.17% of elapsed time. At 10 minutes: 600 × 0.0417 = 25 s out of sync; over a 120-minute film: 120 × 0.0417 = 5 minutes.
• A 23.976 ↔ 24 fps error is only 0.1%: over 2 hours, 7200 × 0.001 = 7.2 s — small, but enough to fail QC late in the reel.

Example 4 — two-liner feasibility and the six-second rule.

• A full two-liner at 42 characters per line = 84 characters. At 17 CPS, t_min = 84 / 17 = 4.94 s, so a full two-liner needs about 5 seconds.
• The classic six-second rule (2 × 37 = 74 characters in 6 s) works out to 74 / 6 = 12.3 CPS ≈ 123 WPM — markedly slower, which is why old TV subtitles feel sparse against OTT subtitles.

Example 5 — episode effort and billing (indicative rates). A 44-minute web episode; measured speech time 32 minutes; average subtitle duration 2.8 s.

• Subtitle events ≈ (32 × 60) / 2.8 = 1920 / 2.8 ≈ 686 subtitles.
• Translation at 120 subtitles/hour → 686 / 120 ≈ 5.7 h; review and simulation ≈ 1.5 h → ≈ 7.2 h total.
• At ₹150 per runtime minute: 44 × 150 = ₹6,600; GST 18% = ₹1,188 → invoice ₹7,788. Effective earning = 6,600 / 7.2 ≈ ₹917 per hour before tax. (Market rates vary widely by language pair and client; treat as an illustration.)

Example 6 — dub line syllable fit. Original English line: 9 syllables in 2.6 s → R = 9 / 2.6 = 3.46 syllables/s.

• Hindi first draft has 15 syllables → needed time = 15 / 3.46 = 4.33 s, i.e. 1.73 s over the loop.
• Fit requires about R × 2.6 ≈ 9 syllables (±10% ⇒ roughly 8–10); the adapter must rewrite, not just trim, and keep any visible /p, b, m/ at the same instants.

Where it is used

• OTT localisation vendors (Prime Focus, Pixelogic-type houses, Ooona-based studios): roles of subtitler, template author, QC/quality checker, language lead, project coordinator for Netflix, Prime Video, JioHotstar, ZEE5, SonyLIV, Aha, Hoichoi.
• Film dubbing studios in Chennai, Hyderabad, Mumbai and Kochi: dialogue adapter, dubbing director, sync assistant, voice artist — the backbone of pan-India releases (Baahubali, KGF, Pushpa, Kantara and the Hindi/Tamil/Telugu dubs of Korean and Japanese content).
• Broadcast and public media: Doordarshan and Prasar Bharati captioning, news voice-over, ISL interpretation; regional news channels running same-language captions.
• Accessibility services: SDH and audio-description writing for theatrical releases under the MIB accessibility standards, and for government and educational video.
• EdTech, e-learning and corporate video: subtitling and dubbing of courses (SWAYAM, NPTEL, Coursera-type catalogues) and training films into Indian languages — large, steady freelance volume.
• Advertising, gaming and AI data work: ad-film dubbing in 8-12 languages, game dialogue localisation, and transcription/QC for speech datasets under Bhashini-type programmes.

Interview questions

• Q: Why is subtitling called constrained translation? A: Because the translation must fit a fixed space and time, stay in sync with the image and shot changes, and coexist with the audible original, so the translator condenses and prioritises rather than rendering everything.
• Q: What does 17 CPS mean and what happens if you exceed it? A: Seventeen characters per second of on-screen time; above it the average viewer cannot finish reading before the subtitle disappears, so information is lost and the image goes unwatched.
• Q: What is spotting, and why must a subtitle not cross a shot change? A: Spotting is fixing each subtitle's in and out timecodes to the speech; a subtitle that straddles a cut makes the eye re-read it, because a cut triggers a fresh visual scan.
• Q: How do SDH and ordinary subtitles differ? A: SDH adds speaker identification and non-speech information (sound effects, music, tone) for viewers who cannot hear the track; ordinary subtitles assume the audio is audible.
• Q: What is an M&E track and why does dubbing depend on it? A: The music-and-effects mix without dialogue; the dubbed voices are laid over it, so without an M&E the studio must rebuild Foley, ambience and crowd from scratch.
• Q: Name the types of synchrony a dubbing adapter works with. A: Phonetic/lip sync on visible consonants and open vowels, isochrony of line length, kinesic sync with gestures, plus character and content sync.
• Q: What is template or pivot translation and what is its risk? A: A timed English master file is translated into many languages; it saves cost but every mistranslation or wrong cue in the template is inherited by all targets, and source-language nuance is lost twice over.
• Q: A client sends an SRT timed to 24 fps but the master is 25 fps. What do you do? A: Conform/retime the file — a 24→25 assumption drifts about 4.17% of elapsed time, roughly 25 seconds by the ten-minute mark, so the whole file must be rescaled, not nudged.
• Q: Does a dubbed Hindi version of a Tamil film need a fresh CBFC certificate? A: Yes — under the Cinematograph Act, 1952, each language version certified for public exhibition is certified separately, and the 2023 amendment's age categories (UA 7+, 13+, 16+) apply to it.
• Q: Which rights must be cleared before subtitling or dubbing a film? A: The translation and adaptation rights under the Copyright Act, 1957 from the rights-holder of the film and of underlying works, including song lyrics; the resulting subtitle or dub script is itself a derivative work whose ownership must be settled in the contract.
• Q: What accessibility duties apply to Indian audiovisual content? A: Section 42 of the RPwD Act, 2016 requires audio description, closed captions and sign-language interpretation in electronic media; the MIB's 2024 accessibility standards require embedded closed captions and audio description for feature films in theatres on a phased timeline, and the IT Rules, 2021 expect OTT publishers to improve accessibility.
• Q: Where does AI help and where does it fail in this work? A: ASR and MT give usable first drafts of transcripts and subtitles and AI voices can serve low-value content, but they misread Indian names, code-mixing and overlapping speech, cannot judge condensation or register, and raise consent issues when a voice is cloned — so human post-editing and QC remain mandatory.

Common mistakes

• Translating the dialogue list fully instead of condensing — a subtitle that is "accurate" at 25 CPS is unreadable and therefore wrong.
• Breaking lines by character count rather than syntax, leaving a stranded article, preposition or auxiliary at a line end.
• Ignoring shot changes and the minimum gap, so subtitles flash or run across cuts.
• Confusing the frame rate or the timecode base (24 vs 23.976 vs 25 fps; frames vs milliseconds) and delivering a file that drifts.
• Counting Indic characters by code point in one tool and by grapheme cluster in another, so the file passes locally and fails client QC.
• In dubbing, writing a semantically perfect line that is 50% too long, or ignoring visible labials and the character's register and age.
• Forgetting the legal layer in exams and interviews: no translation right cleared, no separate certificate for the dubbed version, no accessibility deliverable, no written assignment of the subtitle file.
• Using a plot-spoiling subtitle before the on-screen reveal, or subtitling songs and on-screen text without checking whether forced narratives were commissioned.

Related topics

Translation · Machine translation, post-editing and localisation · Editing and proofreading · Content writing · Media and film studies · Copyright and intellectual property in media