Dialogue-only subtitles vs CC: Netflix's two English tracks, burned

Dialogue-only subtitles show what people say and nothing else. Closed captions for the deaf and hard of hearing (SDH/CC) add speaker names and sound cues such as [door slams]. Netflix now lists both as separate English choices, so a viewer picks the density they want. If you burn text into a video you have to pick for them, and with Sume you can build both versions from one script and render each as its own job.

The facts about Netflix's two tracks below come from its announcement of the new subtitle option, read 2026-10-03.

What are the two English options on Netflix?

Netflix says the language picker on a new title now shows two entries: English, which shows only the spoken dialogue, and English (CC), which includes dialogue plus audio cues like [door slams]. Before this, a viewer who wanted original-language subtitles had to turn on SDH/CC, which also carried cues like [phone buzzing] or [dramatic music swells] and speaker names.

Netflix adds that the dialogue-only option will be available on all new Netflix originals in every language it offers, in addition to SDH/CC, and that viewers can change the size and font of their subtitles. It also reports that nearly half of all viewing hours in the US happen with subtitles or captions on.

How do the two versions differ?

The split is a content decision, not a styling one. The table lists what each version carries, per Netflix's description, and what that means when you author the text yourself.

Which one should a burned-in caption be?

Burned-in text cannot be toggled, so it should match the audience that cannot opt out. For a muted autoplay feed, dialogue only keeps the picture clear. For a clip that deaf or hard-of-hearing viewers must follow, the sound cues carry meaning, and the version with cues is the one to burn. If you serve both audiences, render two files and publish each where it fits, instead of one compromise.

Sound cues should be short and only for sounds that matter. A doorbell that drives the scene earns a cue; footsteps you can see do not.

How do I render both versions from one script?

Standalone video captions accept cues, each with text, start and end in seconds, which skip speech-to-text and burn exactly that copy at those times. Keep one list with a kind field, filter it twice, and submit two jobs. Each accepted job for a clip up to 60 seconds is $0.20 of Sume usage, per the video captions docs. script_text, words, cues and segments are mutually exclusive, so send only cues.

import json

script = [
    {"kind": "speech", "text": "Did you hear that?", "start": 0.4, "end": 1.8},
    {"kind": "sound", "text": "[door slams]", "start": 2.0, "end": 3.0},
    {"kind": "speech", "text": "Someone is here.", "start": 3.2, "end": 4.6},
]

def body(kinds, video_url):
    cues = [{k: c[k] for k in ("text", "start", "end")}
            for c in script if c["kind"] in kinds]
    return {"video_url": video_url, "cues": cues}

url = "https://media.sume.com/artifacts/example/clean.mp4"
print(json.dumps(body({"speech"}, url), indent=1))
print(json.dumps(body({"speech", "sound"}, url), indent=1))

POST each body to /v1/video-captions with its own Idempotency-Key, then poll the job as described in Jobs and results. A silent clip works with cues, since no speech-to-text runs. If a viewer will see the text on a phone, check the line length first; subtitle limit per line covers max_chars.