Editing Auto-Generated Captions

Media has become a foundational tool for communication. However, the media we produce must include accurate closed captions to be useful and accessible to our audiences. Even if we know captions are important, how do we get started adding them? This guide will walk you through one method to produce high-quality captions: beginning with auto-generated captions provided by a video hosting platform and editing them for accuracy.

Table of contents

• Who is this guide for?
• About auto-generated captions
  • When to add
  • Platform-specific resources
• Example: Before and after editing
• Guiding principles
• Proofreading
  • Punctuation
  • Spelling
  • False starts, filler words, repeated words, and more
  • General practices
• Additions
  • Speaker names
  • Crosstalk
  • Added sounds
• Display and formatting
  • Length
  • Line breaks
  • Timing
  • Position
• Other approaches to captioning
• Resources
• Credits

Who is this guide for?

This guide is designed for people new to providing closed captions on video. This guide will help you make your media accessible, even if you don’t have the budget for professional captioning or transcription.

About auto-generated captions

Auto-generated captions, also known as automatic captions or Automatic Speech Recognition (ASR) captions, use machine learning to turn audio into written text. These captions can be accurate. However, they are not accurate enough to meet accessibility guidelines or users’ needs. Auto-generated captions need a human editor to fix mistakes and add elements like speaker names or sound effects to the captions for full accessibility.

When to add

When producing a video, build in time towards the end of your project for editing captions. Be sure that you’ve made all needed edits to your media, then upload it to your video hosting platform of choice (for example, YouTube). If you edit your captions, and then need to change your media, you will have to start over. Your video platform will take some time to generate captions. Once complete, you can edit them as a final step before publishing.

Platform-specific resources

This guide is intended to help you know what to edit and add to auto-generated captions, regardless of the platform you use. For details on how to find and edit auto-generated captions on a specific platform, visit the following resources.

• YouTube
  • Use automatic captioning
  • Edit or remove captions
• Vimeo
  • FAQ: Automatic closed captions
  • How to edit a video’s transcript
• Kaltura
  • Editing Captions

Example: Before and after editing

Compare the closed captions on the two videos below. The first video uses captions automatically generated by YouTube without any human editing. The second video shows the same set of captions after the captions have been edited and enhanced. Try watching the videos without sound. What changes between the two caption tracks make the captions more useable?

Video with unedited captions

Video with edited captions

Guiding principles

When editing captions, best practices and guidelines can help you craft useful and accessible captions. However, all caption editors will have to make judgement calls about what to include and how to describe something. When making decisions, keep these principles in mind.

• Accurate: Errorless captions are the goal for each production.
• Consistent: Uniformity in style and presentation of all captioning features is crucial for viewer understanding.
• Clear: A complete textual representation of the audio, including speaker identification and non-speech information, provides clarity.
• Readable: Captions are displayed with enough time to be read completely, are in synchronization with the audio, and are not obscured by (nor do they obscure) the visual content.
• Equal: Equal access requires that the meaning and intention of the material is completely preserved.

Elements of Quality Captioning from the DCMP

Tip: Don’t let perfect be the enemy of good

Writing captions for the first time may feel overwhelming. As you begin, don’t focus on memorizing all the specific guidelines and rules. Rather, edit with your audience and these five principles to guide you, and you’ll be able to learn the nuances over time.

Proofreading

Punctuation

Depending on the platform used, the auto-generated captions may or may not have punctuation added. Captions should be punctuated to promote readability and understanding. If your captions do not have punctuation, add it as you edit. If your captions have punctuation, make sure that the placement is correct. Occasionally, the caption generator will misinterpret pauses and add punctuation that does not reflect the message being communicated.

We do not always speak in fully-formed, grammatically-correct sentences. Use your best judgement in punctuating captions. In general, simple punctuation and multiple shorter sentences are better.

Tip: Think about display

Remember, only a portion of the sentence will appear on screen at a time. Use parenthesis, colons, and semi-colons sparingly. Your edits should support understanding, not distract.

Spelling

Check the spelling of words, and that the auto-generated captions have the correct word. Some caption editors will have spell check built in, but don’t rely fully on this.

Common errors

Watch for these common errors in your captions.

• Names: Automatic speech recognition may spell names incorrectly or mistake them for other words. Always check names of people, places, and organizations/companies to confirm they are spelled correctly.
• Technical/scientific words: If your media uses technical or specialized language, the caption generator may mistake these terms for more common words. If you are not certain how to spell a term, search for the term or consult a subject matter expert.
• Homonyms: Autogenerated captions may mix up words that sound the same but have different spelling or meanings. For example, accept and except or soar and sore.

Tip: Search for repeated mistakes

If you notice that a mistake is repeated, many caption editors have a find and replace function. For example, if a name appears misspelled the same way throughout captions, you can use Find or Find and Replace to quickly identify and edit all similar mistakes.

What if you can’t tell what a word/sound is?

Sometimes the audio quality or background noises make it difficult to tell what was said. Make a best effort to figure out what the missing word is. Consult a second person (or the speaker) if possible. If the word is still not identified, add [inaudible]. Do not guess.

False starts, filler words, repeated words, and more

Many speakers may use filler words (um, uh) or repeat the same word multiple times in a row while they speak. They may also make a false start in their speech, then start over. These elements do not need to be captured in captions.

• Delete filler words.
• Only use one instance of a repeated or stammered word.
• If the false start is brief, just include the actual start.

Captioning conventions vary on how they approach these situations. Make editorial choices that prioritize understandability and readability of the captions.

General practices

• Numbers: spell out one to ten, use numerals for the rest.
  • Spell out numbers at the beginning of sentence. Thirty people attended the event.
  • Do not mix spelled out numbers and numerals in the same sentence. Choose one method and stick with it.
  • There are many different considerations with numbers. If you want to explore more, visit the DCMP Captioning Key - Numbers
• Capitalization: follow normal capitalization conventions. All caps indicates shouting. Use sparingly.
• Spelling: If a speaker spells out a word, type out the letters with dashes between. For example: H-E-L-L-O
• Websites: Put in the URL as it would appear in a browser, even if the speaker spells it out. www.dcmp.org
• Quoted material: If the speaker is quoting/reading someone else, but a quotation mark at the front of each caption segment. Do not add a closing segment until they are finished.

Example of quoted material

Additions

In addition to fixing errors made in the auto-generated captions, there may be additional information you need to add to the captions to make them useful.

Speaker names

If you have more than one voice speaking, you should capture who is speaking. To do so, add the speaker’s name before the caption segment each time the speaker switches. There are several standard ways to denote a speaker:

(Speaker Name) 
Captioned line of text

or

Speaker Name: Captioned line of text

Whichever method you choose, keep the following points in mind.

• Use a consistent method throughout the video.
• Do not add the speaker information before every caption segment, only when the speaker switches.
• If you do not know a speaker’s name, use an alternative such as (Speaker #1) or (Narrator).

Crosstalk

Crosstalk, when two or more speakers talk at the same time, can be another challenge to capture in captions. There are several approaches.

• If you cannot decipher what is said, use [crosstalk] to note this.
• If the crosstalk is brief enough, include one person’s line right after the other (with speaker names).
• If there is a dominant voice that continues, capture the dominant voice.

Example of capturing speaker names and crosstalk

Added sounds

Auto-generated captions will generally only capture speech. For captions to be equal, non-speech sounds should be included. Other sounds like music or sound effects should be added to captions in brackets. For example [door slamming]. There are several ways you might capture music and sounds in captions.

• If the name and/or artist of the song is known, add this information [in brackets].
• Song lyrics should be captioned. Add a ♪ before each sung caption segment. When singing ends, add two music notes ♪♪ after the last word.
• For background music, write a short description that indicates mood. [gentle piano music].
• Sound effects or environmental sounds can also be capture in brackets, [applause], [cow mooing] or in onomatopoeia Mooooo.
• When relevant, capture the source of a sound. [audience laughter].

Example of music in captions

Do I need to capture every sound?

Use your best judgement with capturing sound. Your goal is to provide an equivalent experience for someone reading captions and someone hearing. If too many sounds are captured, there may not be enough time to read the captions. For example, if an audience member coughs during a speech, that cough may not be important to capture. However, if the speaker stops and reacts to the cough, the cough would be useful information to capture.

Display and formatting

When editing auto-generated captions, many timing and formatting decisions will already be made for you. As you add to and edit captions, keep the following guidelines in mind.

Length

Each caption segment should be one to two lines. Information like speaker name may be placed on a third line. A caption segment should display one to six seconds.

Line breaks

When breaking up lines within a caption segment or between caption segments, put breaks in logical places.

• Keep words that belong together together (for example, modifiers and modified, prepositional phrases, names, etc.)
• Don’t end a sentence and start a new one on the same line.
• Start a new caption segment whenever the speaker switches.

Timing

The caption generator should set up the caption timings for you. You can manually adjust the start and end times as needed. If the timing is off for the entire video, it is better to start again.

Position

Many of the caption generators mentioned in this guide do not have settings for the position of the captions. If you can change this, make sure captions don’t cover up other text. Keep them in a consistent spot.

Tip: Plan for captions

When producing your video, find out in advance where captions will be placed in the video platforms you will use. Make sure nothing appears in this space that will be covered up by captions (such as title cards, etc.) Alternatively, leave black space below the video for captions to display.

Other approaches to captioning

This approach to editing captions is just one method. You may wish to consider alternative approaches based on your budget and workflow.

• Video editing software may have the ability to add or generate captions.
  • Add captions while editing the video. When exporting the video, include them directly on the video so they are always present.
  • Create captions while editing, export a caption file (.srt or .vtt are common file types), and then upload them to your video hosting platform.
• If you have a script or accurate transcript, you can submit the file to your video hosting platform to turn into captions.
• Manually transcribe captions from scratch (not recommended).
• Pay for professional captioning.

Resources

This guide is based on the Described and Captioned Media Program’s Captioning Key. While this guide will get you started in editing quality captions, the list of recommendations is not exhaustive. Consult the Captioning Key for more details.

Credits

Videos used for examples are published by NASA and are in the public domain. Specific videos include:

• Evolution of Galaxies
• NASA Astronauts Celebrate Asian American and Pacific Islander (AAPI) Heritage Month 2021 Advancing Leaders Through Purpose-Driven Service
• NASA in Silicon Valley Live - Halloween Costume and Cosplay Contest
• Happy Birthday Curiosity

This site uses Just the Docs, a documentation theme for Jekyll.

Site created by Kaia Sievert. This work is licensed under a CC BY-NC 4.0 license, which means you are free to reuse the content for non-commercial uses with attribution to the original project.