YouTube expressive captions: ALL CAPS and [sighs], burned in

Expressive captions are a YouTube feature of English automatic captions: they show the intensity of speech in all caps, add expressions of sounds like sighs and gasps, and add noises from the environment. YouTube's help page says it makes captions more immersive. They live on YouTube's own caption track, so they do not travel with your file. To get the same effect in a video you post anywhere, you author the cues yourself and burn them, since Sume's caption job burns exactly the text you pass as cues.

The feature description comes from YouTube Help's Use automatic captioning page, read 2026-10-03.

What does YouTube's page say about expressive captions?

Automatic captions in English support expressive captions. YouTube lists three behaviors: intensity of speech shown in all caps, expressions of sounds such as sighs and gasps, and environmental noise. The same page says automatic captions come from machine learning and can misrepresent speech because of mispronunciations, accents, dialects or background noise, and it encourages creators to add professional captions first and to review the automatic ones.

Two limits matter here. Automatic captions on a video with multiple audio language tracks follow the default language. And a video can fail to get automatic captions at all, for reasons the page lists: processing time, an unsupported language, a very long video, poor sound, a long silence at the start, or overlapping speakers.

How do you reproduce each behavior in a burned-in caption?

Burned-in text has no toggle and no machine-learning layer, so each behavior becomes a writing choice. The table maps them.

Is shouting in caps a good idea?

In moderation. All caps for one shouted word is a clear signal; all caps for whole lines is hard to read and reads as noise. Keep sound cues brief and put them in their own cue, so the spoken line stays readable. Sume caption styles also carry their own emphasis: a spoken word can shift weight or color. Check one frame of the render to see whether your capitals and the style's emphasis are doubling up.

What does the request look like?

Each cue needs text, start and end in seconds. Cues skip speech-to-text, so they also work on a silent clip. cues cannot be combined with script_text, words or segments in the same job.

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: expressive-cues-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/example/clean.mp4",
    "cues": [
      {"text": "[sighs]", "start": 0.2, "end": 1.0},
      {"text": "I said STOP.", "start": 1.2, "end": 2.6},
      {"text": "[traffic]", "start": 2.8, "end": 4.0}
    ]
  }'

When is YouTube's own track enough?

If the video lives on YouTube and a viewer can turn captions on, the platform track may be all you need, as long as you review it. Burn captions when the clip is reposted, when the video will autoplay muted on another platform, or when you want one exact version everywhere. A job for a clip up to 60 seconds is $0.20, per the video captions docs. Restyling the same video later with source_caption_id reuses the stored timings and runs no second speech-to-text, which is handy for trying a calmer style on the same timings.