Subtitle Tools Pro

Subtitle Readability: CPS, Line Length and Reading Time

A subtitle that is on screen at the right moment can still be impossible to read. Readability has three measurable limits, and every one of them is routinely broken.

10 min read · Updated 2026-11-16

Timing and readability are separate problems and get conflated constantly. A file can be perfectly synchronised — every line appearing exactly when the sound does — and still be unwatchable, because the viewer cannot physically read the text before it disappears. That is not a timing failure. It is a reading-rate failure, and it has its own numbers.

There are three of them: characters per second, line length, and how long a cue stays on screen. Each has an industry limit, each is broken more often than anyone admits, and each fails in a different way. This guide is about what the limits actually are, where they come from, and how to find the cues that break them.

Characters per second

CPS is the reading speed a cue demands. Divide the number of characters by the time it is on screen. A cue with 42 characters shown for 3 seconds is 14 CPS; the same cue shown for 2 seconds is 21 CPS.

The numbers that matter, from the sources that actually publish them:

SourceLimitNotes
Netflix (English adult)20 CPSHard limit in the delivery spec
Netflix (English children)17 CPSLower for younger readers
BBC (adult)15-16 CPSLower than Netflix
EBU / European broadcast15-17 CPSVaries by country
Common practice in subtitling software17-21 CPSWarning thresholds, not hard limits

The spread is wider than people expect — BBC and Netflix differ by around 25% — because reading speed depends on the audience and on how much of the original dialogue is being compressed. What the sources agree on is the shape: below roughly 12 CPS almost nobody struggles, above roughly 21 almost everybody does, and in between it depends.

Count characters including spaces. Some tools count only non-whitespace characters, which understates the real rate by 15-20% on ordinary English prose. A file that scores 17 CPS under the stricter count is closer to 20 CPS in the Netflix sense — right at the limit, not comfortably under it.

Line length

Two lines of up to 42 characters is the standard, and it comes from the same place as the CPS numbers: what fits on screen at a legible size on a television. Netflix allows 42 characters per line. The BBC allows 37. Most subtitling software warns at 42 and refuses at 50 or 60.

Line length fails differently from CPS. A cue with one very long line is not necessarily too fast to read — it might be on screen for five seconds — but the reader has to sweep their eyes across the whole width, which is slower than reading a compact block, and on a narrow screen the line gets truncated or wrapped by the player into something the author never designed.

The related rule is the line count: two lines maximum. A three-line cue occupies too much of the picture and, in practice, is a sign that the cue should have been split into two.

MetricIdealWarningBroken
Characters per line≤ 4243-50> 50
Lines per cue≤ 23> 3
Characters per second≤ 1718-20> 20
Cue duration1-7 s0.5-1 s or 7-10 s< 0.5 s or > 10 s

How long a cue stays up

Duration is bounded on both sides, and the lower bound is the one people forget. A cue under half a second is not readable no matter how few characters it contains — the eye needs time to find it and the brain needs time to register it. A cue over ten seconds outstays its welcome and starts to feel like part of the picture rather than text, so the viewer stops reading it and then misses the next one.

The practical range is one to seven seconds. Below one second, the cue is almost certainly a timing artefact — a line that got squeezed by a neighbouring cue — rather than a deliberate choice.

There is a second, subtler problem here: the gap between cues. Two cues separated by less than about 100 milliseconds read as one continuous block of text that flickers, because the eye does not register the change as a break. If a subtitle file looks like it is flashing rather than changing, tight gaps are the reason.

Why the average hides the problem

This is the most useful thing to understand about readability scores. A file with one unreadable cue and forty comfortable ones has a perfectly healthy average, and the average tells you nothing about the cue that will make a viewer rewind.

Concretely, a three-cue file:

CueTextDurationCPS
1This line is forty two characters long ok2.0 s17.0
2Too fast0.3 s23.3
3Stays far too long on screen for anyone to enjoy reading15.0 s3.1

The average CPS across this file is 5.0 — far below any limit, and completely misleading. The maximum is 23.3, above the Netflix hard limit. Cue 2 is unreadable; cue 3 is on screen for fifteen seconds. A report that only shows the average says this file is fine. It is not.

When you look at a readability report, read the maximum and the per-cue list before the average. The average is useful for spotting a file that is uniformly too fast, which is a real and different problem, but it will never find the one cue that ruins the experience.

Why the same numbers do not work for every language

CPS assumes a character-based writing system where a character roughly maps to a syllable or a phoneme. That holds for Latin scripts. It does not hold for Chinese or Japanese, where a single character carries far more meaning and is read as a unit rather than assembled letter by letter.

The practical consequence: a Chinese subtitle line of 15 characters carries roughly the information of an English line of 40 characters, and both take about the same time to read. Applying the English CPS limit of 20 to Chinese text would flag almost every correct cue as too fast, because the same information arrives in a third of the characters.

In practice, Chinese and Japanese subtitles are limited by line length — around 13 to 16 characters per line, two lines maximum — rather than by CPS. If you are checking a CJK file with an English-configured tool, expect false positives and read the per-cue detail rather than the score.

What to do about a cue that is too fast

There are only four honest options, and they are not interchangeable:

  1. Extend the cue. Move the out-time later, if the next cue leaves room. This is the least destructive fix and it is usually available more often than people assume.
  2. Split the cue. If the text is genuinely too long for its slot, divide it across two cues with a small gap. The words stay; the reading burden halves.
  3. Shorten the text. Condense the line — drop filler words, use a contraction, cut a redundant name. This changes the translation, so it is a decision for whoever owns the subtitles, not a mechanical fix.
  4. Shift the timing. If the cue is fast because the speech itself is fast, the honest answer is often that the dialogue cannot be fully subtitled at that speed, and something has to give.

What does not work is lowering the threshold until the file passes. A report that says "no problems" because the limits were relaxed has told you nothing.

Checking a file yourself

You do not need special software to do a first pass. The four checks below catch most real problems and take about a minute on a file of normal length:

  1. Sort the cues by CPS descending and look at the top ten. Those are the ones a viewer will have to rewind.
  2. Look for any cue under 0.5 seconds. Almost all of them are timing artefacts.
  3. Look for any cue over 7 seconds. Long cues usually mean two cues were merged by mistake.
  4. Count the characters in the longest line. Anything over 50 will be truncated on some players.

Try it on your own file — no upload, nothing leaves your browser.

Open the Subtitle Quality Checker - CPS, Overlap & Duration Report →

If the file passes those four checks and still looks wrong on screen, the problem is more likely to be timing than readability — a different diagnosis with a different fix.