Put the punctuation back

Why captions lose their punctuation

Speech recognition hears words, not commas. Punctuation is a second guess made on top of the transcript, and in fast or casual speech many tools make that guess rarely or not at all. The result is a wall of lowercase words that is hard to follow when it flashes past a few at a time.

Missing full stops also hurt caption breaks. In Cuecaps a full stop, question mark or exclamation mark always ends a caption, so filling them in first gives every caption a cleaner ending.

When Cuecaps adds a mark, and when it leaves things alone

• It steps in when the transcript is almost bare, with fewer than one word in thirty carrying punctuation, or inside any stretch of 15 or more words without a single mark.
• It adds a comma, full stop, question mark or exclamation mark only where the model is fairly sure one belongs.
• A word that already ends in punctuation is never changed.
• Capitals are filled in only when the transcript has none at all, and the word after an added full stop, question mark or exclamation mark starts with a capital.

The same on-device caption model does this as part of picking caption breaks, right after Find words. On hand-marked real clips it scored 0.82 F1 for punctuation and 0.74 for capitals. That is good enough to save most of the typing, but read the result once, especially questions.

Finishing the job by hand

1. Open your video in the studio and run Find words.
2. Read through the transcript (the Words tab on a phone) and edit any word to add or remove a mark.
3. If you prefer captions without commas and full stops on screen, turn on Hide punctuation except ? and ! in the Text tab. The sentence structure still guides where captions break.
4. Export the MP4 for social, or the SRT and VTT files for YouTube and other players.

A serif or typewriter style makes the restored punctuation part of the look:

Questions