Information Technology Systems and Services.
ITSS home

Improve Accessibility: Correct YouTube Auto-Captions

YouTube auto-captions are often such poor quality that content is not accurately communicated to people who depend on captions such as people who are Deaf and hard of hearing. Auto-captions should be corrected to be precise and provide equal access. A redeeming aspect of auto-captions is that they can be used as a starting point for captioning your own videos. The Multimedia Hub is available to assist you in providing videos inclusive to all.

Auto-Caption Inaccuracies

Icon: YouTube Closed Captions The ability to use voice recognition technology to automatically generate captions for video would significantly reduce the cost of captions. It holds promise. However, YouTube's automatic captioning currently does not provide a high enough accuracy rate to be considered usable and compliant with university accessibility standards. The only way to know for sure if automatic captions are accurate is to activate them and play the video. Never assume that automatically generated captions are acceptable.

Auto captioning often produces grossly inaccurate results. If you haven't watched any of Rhett McLaughlin and Link Neal's Caption Fail Videos, it is highly recommended not only for the comic and entertainment value, but also as a demonstration of the issues with automated captioning. YouTube automatic captions typically provides about 60-70% accuracy, which means that 1 in 3 words can be wrong. This accuracy rate will be improved with good audio quality and simple content, but worsens when there is background noise, accents, or multi-syllable words.

Accurate captions can not only keep your videos from turning into a joke, they can also help UMD avoid being put at risk for legal action. In 2013 the University of Maryland was sued for uncaptioned videos and other media on their athletic department's website. Recently Harvard and MIT were both sued by advocates for people who are Deaf for violating anti-discrimination laws by failing to provide captioning for their online media. The National Association of the Deaf (NAD) accused the universities of violating the Americans with Disabilities Act, stating that the materials were either unintelligibly captioned or completely uncaptioned, making it impossible for viewers who are Deaf and hard of hearing to understand the content. "Worse still," said Timothy Fox, Executive Director of the Civil Rights Education and Enforcement Center and a lawyer for the plaintiffs in both cases, "a sampling of the videos available illustrates the problem with inaccurate captioning, making them confusing and sometimes completely unintelligible." 

Accurate Captions Help Ensure Inclusiveness

Accurate captioning (of at least a 99% accuracy rate) is the only way to ensure that people who are Deaf and hard of hearing can understand audio content. In fact, most accessibility advocates would argue that automatic generated captions are actually detrimental to accessibility.

Accurate captions aid more than people with hearing loss. As 3Play Media points out in the article, 80% of People Who Use Closed Captions Are Not Hard of Hearing:

Viewers who know English as a second language often benefit from closed captions, because it makes it easier to follow along with spoken content that is not as familiar to them. Closed captions help with comprehension: dialogue that is spoken very quickly benefits from captioning, as does dialogue with accents, mumbling, background noise, or complicated/esoteric subject matter. For video that is published online, closed captions increase viewer retention and user engagement, as well as search engine optimization. Captions allow viewers to watch videos in sound-sensitive environments like offices and libraries.

Using Auto-Captions as a Starting Point

If you are the video owner, YouTube's auto-captions can be very useful as a starting point for providing captions. Please note: You can only edit your own Youtube videos. Certain videos that are uploaded to YouTube are good candidates for using their machine generated captions as a base. If the voices are clear, speaking unaccented English, and there is no music and minimal background noise, auto-captions can get you started.

  1. Start YouTube. It creates captions, but they won't be accurate enough on their own.
  2. Go back and manually edit the text. This way, you just need to fix the problem spots rather than typing in the entire text.
  3. Add appropriate sentence delineation (punctuation and capitalization.)
  4. Provide information about significant sound effects. For instance add descriptions of sound in square brackets (such as [music] or [laughter]) to help people understand what is happening.
  5. Make sure the captioning is in sync with the audio.

Google's Edit Captions documentation provides instructions. The following 6 minute video, "How to Edit YouTube's Auto-Generated Captions," by Cincinnati State provides a walk through the process.

Need Help?

The Multimedia Hub is available to assist the UMD community in providing videos inclusive to all. The Hub is happy to work with departments that have YouTube videos posted. For instance they can train students, faculty, and staff members how to edit automatic YouTube closed captioning to make it accessible.

Further Information