Captions for the town meeting video
Picture a town clerk who posts council meeting videos to the town website every month. The videos are recorded on a camera at the back of the room, uploaded to YouTube as unlisted, and embedded on the site. YouTube generates automatic captions. The clerk assumes that satisfies the accessibility requirement. A resident who is deaf submits a complaint. The auto-captions render the mayor's name as "male Williams," the street name "Quapaw" as "cow paw," and the budget figure "$40,000" as "farting thousand dollars." The town is now in violation of Title II of the Americans with Disabilities Act. The clerk has sixty days to fix six months of videos.
Recorded government meetings are public records. If a resident cannot attend in person, the video is how they participate. If the resident cannot hear, captions are not optional. Auto-captions are better than nothing, but they are not compliant. The standard is that captions must be accurate, synchronized, complete, and properly formatted. Auto-captions fail accuracy the moment someone says a local name, a technical term, or a number. They fail synchronization when multiple people speak or when there is background noise. A town that posts videos with uncorrected auto-captions is publishing a public record that a portion of the public cannot use.
What auto-captions get wrong
Automatic captions are generated by a speech recognition model. The model is trained on common words and phrases. It does not know the names of your streets, your board members, or your local businesses. It guesses based on what sounds similar. "Piggott" becomes "pigot." "Corning" becomes "corning." "Rector" becomes "rector" if you are lucky and "wrecker" if you are not. The model struggles with accents, with speakers who talk quickly, and with speakers who mumble. It cannot tell the difference between "$15,000" and "$50,000" if the audio is unclear.
Auto-captions also fail on structure. They do not add speaker labels. They do not paragraph. They run every sentence together in a single block. A reader cannot tell when a new person is talking or when the topic has changed. The captions are technically present, but they are not usable.
Corrected captions versus transcripts
Corrected captions are auto-captions that a person has reviewed and fixed. You download the auto-caption file from YouTube, open it in a text editor or caption editor, play the video, and correct the errors as you listen. You fix names, numbers, and garbled phrases. You add punctuation. You adjust timing if a caption appears too early or too late. When you are finished, you upload the corrected file. The video now has accurate captions.
A transcript is a separate text document that contains everything that was said. It includes speaker labels, timestamps, and paragraph breaks. A transcript is easier to search and easier to read than captions. Some states require transcripts in addition to captions. Some require transcripts only for meetings where a significant decision was made. Check your state's open meetings law and your ADA coordinator.
Both corrected captions and transcripts take time. A one-hour meeting takes between two and four hours to caption accurately, depending on audio quality and how many speakers overlap. A transcript takes slightly less time because you do not have to adjust timing codes. The work cannot be automated. A person has to listen and compare.
A workflow that fits a small office
The clerk's office does not have four hours per meeting to spend on captions. The workflow has to be sustainable or it will not happen. Here is the version that works.
Post the video with auto-captions immediately. Add a note at the top of the page: "Automatic captions are available now. Corrected captions will be posted within ten business days." That gives residents access to the video right away and sets an expectation for when the accurate version will be ready.
Download the auto-caption file as soon as the video is uploaded. Assign the correction work to whoever has the best combination of time and familiarity with the content. That is usually the clerk, but it can be an assistant or a volunteer if they were present at the meeting. Correct the file in segments. You do not have to finish it in one sitting. Save progress and return to it.
When the corrected file is ready, upload it to YouTube and remove the note from the page. If your state or policy requires a transcript, generate it from the corrected captions using a caption-to-transcript converter or by copying the text into a document and adding speaker labels manually.
Keep a log of which meetings have corrected captions and which are still auto-only. If someone requests an accessible version of an older video, prioritize that one.
Where this fits
Most of the towns I work with host their videos on YouTube and embed them on the town site. I set up the page template, make sure the video player is keyboard accessible, and confirm that captions display correctly. The captioning work itself is done by the clerk or contracted to a transcription service. If the town uses a dedicated video hosting platform like Granicus or Swagit, those platforms include caption correction tools and sometimes offer captioning as a paid add-on. I help evaluate those options and migrate the video library when a town switches platforms. The goal is a process the office can maintain without waiting on a vendor every month.