Subtitle reading speed: check 20 characters per second before burning

The reading speed limit in Netflix's English (USA) Timed Text Style Guide is up to 20 characters per second for adult programs and up to 17 for children's programs. To check your own subtitles, divide each cue's character count by its duration and flag anything above the limit. Sume's caption endpoint accepts authored cues with text, start and end, so you can run that check on the cues before you pay for a render.

The guide was read on 2026-10-02. It does not define how to count characters, so the script below counts every character in the cue text, spaces included; that is a stricter reading than ignoring spaces, and you should adjust it to the rule your delivery spec names.

What are the limits, and where do they apply?

The Netflix page lists reading speed as rule I.14 of its English (USA) guide. It is a delivery rule for Netflix timed text, not a Sume limit: the Sume docs set no reading speed, and a caption job burns whatever cues you send.

The table puts the two limits next to the Sume fields you would use to stay under them.

How do I check a cue list before rendering?

Run the check on the cue list you plan to send. A cue over the limit needs one of three fixes: split the sentence into two cues, shorten the text, or extend the end time into the gap that follows it. Translations are the usual offenders, because the target-language line is often longer than the source line it replaces.

cues = [
    {"text": "Welcome back to the studio.", "start": 0.0, "end": 1.6},
    {"text": "Today we are dubbing one clip into three languages.", "start": 1.6, "end": 3.0},
    {"text": "Check every cue first.", "start": 3.0, "end": 4.2},
]
LIMIT = 20  # characters per second, adult programs

for cue in cues:
    seconds = cue["end"] - cue["start"]
    rate = len(cue["text"]) / seconds
    flag = "OVER" if rate > LIMIT else "ok"
    print(f"{rate:5.1f} cps {flag}  {cue['text']}")

How do I send the cues once they pass?

Send them as cues (or segments) on POST /v1/video-captions. That is the authored overlay path: it skips speech-to-text, and script_text, words, cues and segments are mutually exclusive. A silent clip is the other case this serves; without cues it fails with caption_no_speech.

Cue times are seconds from the start of the video. If the captions drift against the picture, see how to sync subtitles with video.

What should I do next?

Keep the limit in one constant, run the check on every language you ship, and re-time before you render rather than after. A restyle through source_caption_id reuses the word timings but is still a render, so it is billed like one.