Video Transcription

Video Transcription is useful for improving access to video content in searches. Every video OpenAir produces from a mission for documenting progress, informing stake-holders, or otherwise disseminating knowledge beyond OpenAir, should be transcribed and recorded under the appropriate mission subcategory.

Some topics and videos are too long for many visitors. We need to create summaries with just enough detail to make them helpful, but faster. Text summaries are also easier to search.

The short version, the TLDR (too long, didn’t read - or didn’t watch) could be things like:

  • one-minute trailer (to make it a YouTube-shorts-eligible video)
  • a 5000-character summary text (to meet the YouTube summary limit)
  • an abbreviated transcript with some time-stamps (to facilitate deeper dives into the video content)

All of these are potentially useful to some consumers of video content, and all of them need to be searchable with text keywords or hash-tags. The processes for making these summaries are elaborated:

<how to edit a long video into a highlight/trailer?>

How to create an abbreviated transcript and summary of a video:

  1. If the video is not on YouTube, consider posting it there, because YouTube has an automatic voice-to-text transcription feature: providing an automatic transcription with time stamping. It happens behind the scenes automatically. The image below shows a “This is CDR” video transcript example.
  • You display the transcription by pressing the “…” circled in yellow, then the transcript appears on the right.
  • You can search it and also select a section and it will jump to that part of the video.
  • Time stamps are available for convenient correlation. You can also select text for copy and paste - though the format of the transcript is being actively developed as of December 2022, and how to copy is not the same from one time to the next.

    In December 2022, YouTube’s transcription is not “smart”: it transcribes the best sound match to each word so it sometimes make a word-salad, especially if there are background noises or if the speaker has an accent on which the AI was not trained, or if the words are technical or acronyms. Every “um”, “uh”, “ah”, and other aberrant utterance is captured - consider word-searching and delete these filler vocalizations. It is essential to review that the transcript makes sense:
  1. Review the video or transcription minute-by-minute and condense the contents into well-formed complete sentences with aberrations deleted - this is a “cleaned up transcript”. In December 2022, Google’s ChatGPT AI was available on a trial basis and could create such summaries from blocks of raw transcription up to a few minutes in length if they did not contain topic shifts. A variety of other online summarizing tools purported to have comparable abilities, though none appeared satisfactory. All made it necessary to summarize the transcript in pieces. Whether making a condensed transcript or automating it, retain timestamps every several minutes when there is a topic shift or change of speaker. If there are critical visual aids in the video, consider including some screenshot clips in the clean transcript. Furthermore, ChatGPT is adept at language, but not cognizant of factual relationships - it is entirely possible for current AI models to generate sound-good transcripts that do not actually reflect the underlying content. Use with caution!

  2. From a long-form summary of the transcript, collect the main several points of the video. Automated summarizer software does not seem effective at this stage. Each point-statement should be a complete sentence and the main points can often be presented as bullet-items. Fortunately, video creators often make a brief summary of their video if they post it on YouTube, but they may not include keywords or hashtags that would aid topic searching, so in whatever way the main points are presented, be sure to include useful search terms in the text. Do not report names or affiliations at this stage, unless they are essential to the focus of the video.

Guidelines for Episode Summary:

  • Start with what is already in the YouTube description. Edit if needed.
  • Review YouTube comments and any surprises. Raise major issues with Chris Neidl.
  • Provide a description of the video content with the main points. Use keywords.
  • Be journalistic – avoid using your opinions, and if the speaker states an opinion, identify it as an opinion, eg. “according to Dr. xxx”.
  • Insert key highlights or quotes if appropriate - no more than 3-4 of these. Time stamp them when possible.
  • Have someone else review the summary for clarity and final editing. The writing quality reflects on everyone at OpenAir.
  • The total description must be 5000 characters or less (around 800 words). Be brief.