Tanglish Subtitles: Tamil Captions in English Letters

Tamil creators publishing to Reels and Shorts keep running into the same wall. The captioning tool hears Tamil correctly, and then writes it in Tamil script, which is not what most of the audience reads in a caption. The alternative most people try, forcing an English transcription, produces nonsense, because the model is trying to find English words in Tamil speech. What is wanted is the third thing: Tamil words spelled in Latin letters, the way people actually type to each other.

Why the mainstream tools do not give you this

This is not a gap in accuracy. Tools like VEED, Kapwing, Submagic and HappyScribe detect Tamil well and then output Tamil script, because that is the correct rendering of the language and the obvious thing for a subtitle file to contain. General transcription models pointed at English do the opposite and garble the audio into English words that were never said. Neither is wrong, exactly. Neither is Tanglish.

Romanising makes every line longer, and that breaks your line limits

This is the part that catches people out, and it is measurable. Tamil script is compact: one akshara carries a whole syllable. Latin letters spell that syllable out. The same words get substantially longer:

Tamil script      characters   romanised          characters
வணக்கம்                    5   vanakkam                    8
நன்றி                      3   nandri                      6
சாப்பிட்டீங்களா             8   saapitteengala             14

Roughly 1.5 to 2 times longer, consistently. Applied to a whole caption line, that is the difference between a line that fits and one that does not:

வணக்கம் நண்பர்களே இந்த வீடியோவில் நான் சொல்றேன்
  30 characters, fits a 32-character vertical line

vanakkam nanbargale indha videovil naan solren
  46 characters, does not

So a caption limit that was comfortable in Tamil script overflows the moment you switch to Tanglish, and the player wraps it somewhere you did not choose. Plan for shorter sentences rather than the same sentences in a different alphabet. Where the break should land once a line has to split is covered in caption line breaks for short-form video.

A second, invisible version of the same problem

That Tamil sentence is 30 characters to a reader and 47 UTF-16 code units to a computer. Most subtitle tools measure line length with a plain string length, so they see 47 and wrap a line that was never too long. If you work in both scripts, you will meet this in the Tamil-script file and not in the Tanglish one, which makes it look like the tool is behaving randomly. It is explained in subtitle character limits.

There is no standard spelling, so pick one and hold it

Tamil in Latin letters has no official orthography. Long vowels and retroflex consonants have no settled Latin spelling, so the same spoken word can reasonably be written several ways, and an automatic transcript will not be consistent about which one it picks. Across a three-minute video the same word can appear two or three ways, which reads as carelessness even though every spelling is defensible.

This is worth one pass before export rather than a judgement call per line. The subtitle spelling checker groups words that are near-identical across a file and lets you settle on one, so the decision is made once for the whole track instead of forty times.

Label the file for what it is

A Tanglish track is not Tamil and it is not English, and tagging it as either one misleads whatever consumes it next. The script subtag exists for exactly this: ta-Latn is Tamil written in Latin script, the same wayur-Latn is Roman Urdu. The reasoning, and where the tag actually goes in a file, is in what language a Roman-script subtitle file is.

A short checklist

1. Generate the caption draft, then confirm it is romanised and not Tamil script.
2. Expect lines to run 1.5 to 2 times longer than the same words in Tamil script.
3. Re-wrap for the vertical frame, nearer 32 characters than 42.
4. Settle repeated words on one spelling across the whole file.
5. Tag the track ta-Latn where the platform lets you.

AutoCaption produces an editable caption track before export, which is what makes the spelling and line-break passes above possible at all. Start in the caption generator, or use the Tamil subtitle generator when Tamil script is what the audience should read.