There is no single character-per-second (CPS) limit that makes every subtitle readable. CPS measures how much text appears per second, but the applicable cap depends on the platform’s specification, language and audience—and a compliant rate cannot fix bad timing, awkward line breaks or incorrect words. AI subtitle workflows can stumble at any of those separate tasks.
What CPS means in subtitles
Characters per second, or CPS, is the number of characters displayed divided by the time a subtitle remains on screen. If a subtitle contains 60 counted characters and is displayed for 3 seconds, its rate is 20 CPS.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Blackmagic Design USB Davinci Resolve Editor Keyboard | $669.00 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
Counting conventions matter. Netflix says its software counts spaces and punctuation toward both the character count and reading-speed limit. It calculates speed for each subtitle event as well as the file average, so an acceptable average can conceal an individual subtitle that exceeds a limit. See Netflix’s subtitle templates and its reading-speed measurement explanation.
What is a good subtitle reading speed?
Use the specification for the platform, locale and audience you are preparing subtitles for. Two Netflix references illustrate why “the CPS limit” is not one universal number:
#1 Best Overall
- Designed for professional editors who need to work faster and turn over quickly
- Designed for DaVinci Resolve 16
- Integrated search wheel integrated directly into the keyboard
| Guidance | Adults | Children |
|---|---|---|
| Netflix general Subtitle Templates guidance | Up to 17 CPS | Up to 15 CPS |
| Netflix English (USA) Timed Text Style Guide | Up to 20 CPS | Up to 17 CPS |
These are production limits in different Netflix guidance, not competing measurements of a universal human reading threshold. Check the guide that governs the delivery rather than borrowing a cap from another locale or template. The cited pages are current platform guidance and may change.
Some guidance expresses reading pace in words per minute rather than CPS. Those figures should not be converted into a definitive CPS cap: average word length, spaces, punctuation and writing systems differ by language. A BBC guidance result reports 160–180 words per minute and says pace can be adjusted to the program, text editing and shot synchronization, but the surfaced page is not hosted on a BBC domain. That figure should not be treated as verified official BBC policy without an authoritative BBC source. The distinction between a general pacing reference and a platform delivery specification remains important.
Why an acceptable CPS can still be hard to read
CPS describes text volume over time. It does not show whether a subtitle is phrased naturally, synchronized to speech, split at a sensible point or readable alongside the action. A short, dense sentence can be harder to process than a longer but familiar phrase at the same rate. A line break that separates a phrase awkwardly can also slow comprehension without changing the CPS.
Free tools Windows power users keep installed
One-click scans. No signup required.
Timing has to work with speech and picture. Netflix’s timing guidance notes that a shot change can take precedence over extending a subtitle, so text may need editing to fit. Guidance surfaced for the BBC likewise treats speech pace, editing and shot synchronization as relevant editorial choices. In other words, keeping a subtitle on screen longer is not always available or appropriate.
Dialogue may also compete with important on-screen text. Netflix advises prioritizing the most plot-relevant message rather than cramming everything into the subtitle or severely truncating text. For dialogue-heavy material, timing changes and selective editing may be needed together. The right solution is an editorial decision, not just lowering a number in a validator.
Why AI subtitles can miss the mark
“AI subtitles” can describe several different stages: speech recognition to produce a transcript, machine translation to produce another language, and automated timing or segmentation to decide when subtitles appear and where lines break. Success at one stage does not guarantee success at the others. A plausible transcript, for example, may still be badly timed or divided into unreadable units.
Speech recognition and subtitle timing
A 2024 study tested two automatic speech-recognition tools on English and Italian material, including a British talk show and a US film dubbed into Italian. The authors reported shortcomings against the industry expectation they assessed, particularly for Italian, and found segmentation and timing relatively poor in both languages. They concluded that substantial customization and human post-editing were needed for broadcast-ready subtitles. These findings describe the tools and material in that study, not every current system. Read the 2024 ASR study.
Machine translation and meaning
A 2024 Chinese-to-English movie-subtitle case study examined output from Google Neural Machine Translation. It reported errors in approximately a quarter of the subtitles in its collected film sample. Functional-equivalence errors were most common, followed by acceptability and readability issues. The authors identified problems involving meaning, idiomatic language, grammar, line length, proper nouns, culturally specific expressions and incomplete sentences. That rate belongs to this study’s sample and language pair; it is not an industry-wide error rate for AI subtitles. Read the 2024 subtitle-translation case study.
Post-editing can leave structural problems
A 2022 comparison of Swedish subtitle samples found that the post-edited sample from the 2020s was faster and more oral, but less cohesive and complete, less meticulous in punctuation and line breaks, and lower in the quality areas assessed than the 2010s sample. This is a result from a particular sample and study design—not proof that machine translation always reduces quality. Read the Swedish subtitle comparison.
A 2020 process study involving 12 professional subtitlers also identified segmentation and timing as important issues in machine-translation post-editing. Correcting words alone is not enough when text must be divided and synchronized for viewing. Read the post-editing process study.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to review an AI subtitle file
Review the output against the intended platform guide and the actual audio and picture. These checks cover distinct failure modes documented across the cited guidance and studies; no single study used this exact checklist.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Check transcription against the audio. Correct missed, substituted or invented words, including names and overlapping speech.
- Check translation meaning in context. Confirm that idioms, cultural references, proper nouns and the speaker’s intent survive translation.
- Check reading speed under the governing rules. Apply the intended guide’s CPS cap and counting convention. Inspect each subtitle event, not just the file average.
- Check segmentation and line breaks. Keep phrases together where possible and avoid divisions that make the viewer reconstruct the sentence.
- Check timing against dialogue and shot changes. Ensure subtitles appear and disappear at useful moments without conflicting with the image or speech.
- Check completeness and consistency across the scene. Look for omissions, inconsistent names or terminology, and shifts in how speakers are represented.
When comparing two subtitle versions, record the applicable cap and counting method, language and audience, then assess transcription or translation accuracy, segmentation, synchronization, completeness and readability. Average CPS alone cannot show whether individual events fit or whether the subtitles work with the scene.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
