There are two requirements for captions. Live, and prerecorded. If you have a YouTube live stream that stays up after the livestream, you will have to deal with both of these. You’ll have the live captions during the livestream. Then after the live, you’ll need to handle the requirements for prerecorded captions.
Caption rules
- Captions should appear in simple black & white. Properly uploaded closed captions will handle this by default.
- Captions should not be colored.
- Captions should not highlight the spoken word in any way. (sometimes called karaoke captions)
- Captions should be plain, boring, easy to read, and easy to ignore.
- Closed captions are strongly preferred by audiences. This allows people to choose their captions. This also makes correcting any typos easy to do, and allows for captions to be translated into other languages.
- Open captions always appear and are burned directly into the video. These can be used for some shorts on social media, but is generally discouraged.
Prep your speakers in advance for the best captions
- Encourage speakers to speak clearly and use proper enunciation of words.
- Avoid any “um” or filler words. Pauses are much preferable to filler words.
- Focus on the timing and tempo of speech. Avoid rushing or speaking too slowly. AI transcription is much more reliable at a typical speech rate (120 to 150 words per minute).
- Keep speed consistent when reading content (most people speed up when reading content).
- Avoid slang, it gets transcribed very poorly. Slang does not translate well, and even between countries that speak the same language can be very confusing. For example, “boot sale” means a sale on footwear to people in the US. In other countries, it means a sale out of the trunk of a car.
Software for captions
I use HappyScribe to create transcripts. It allows for easy transcript creation and also can create captions from the same upload. I get about 95% to 98% accuracy with it. I do have significant speech training.
YouTube creates auto captions. These are usually of poor quality. They do not use punctuation and frequently use the wrong word. Using only YouTube automatic captions fails accessibility criteria. If you use them, you must human correct them.
Kapwing, Descript, Davinci Resolve, and other video editing platforms can frequently create transcripts & caption files.
Software can get you to mostly correct captions. Captions must be human reviewed and corrected for accuracy.
Live Captions
Livestreams are required to have captions.
Best practice is to have a live human captioner who is accurate.
Most people don’t have the budget for that, so using autocaptions is acceptable as long as there are not too many errors. Error rate will depend on the caption service used, and the speech patterns of the speaker. Those with heavy accents or who don’t have expected enunciation will have poor results with autocaptions.
The best example for live caption is watching a news broadcast with captions on. You’ll see words appear one at a time, or a couple of words at a time. You’ll see they run several lines of text, frequently 3 or more. You’ll notice they tend to miss a lot of punctuation, and are frequently all caps.
Best practice for live captions is to have words appear as they are spoken. When possible, use sentence case for capitalization.
Prerecorded Captions
Prerecorded media (including live streams that are available after the live stream) needs to have corrected captions with proper punctuation and sentence case.
All captions for prerecorded media must be human reviewed.
If there are multiple speakers, then speaker changes need to be clear.
- When the camera clearly identifies the speaker, no additional information is needed.
- When the camera does not clearly identify the speaker, you need to use the person’s name in brackets (ie: [Gen]) to identify the speaker changes. Each change of speaker needs to be identified.
Captions should appear in 1-2 line groups. This means a whole sentence or a significant part of a sentence appears at once.
Do not use individually appearing words for captions in prerecorded media.
Except in a few cases for social media shorts, captions should be closed captions (a .srt file).
- For social media shorts, open captions are ok.
- Closed captions can be corrected in the future (we’ve all made mistakes).
- Closed captions can be translated into other languages.