Sunday, April 16, 2017

Automated Captioning - Is It Good Enough?

Back in 2011 I made quite a few screen capture movies with my voice over to narrate what's going on and I captioned all of them.  This amounted to:

  1. ripping the audio track out of the video,
  2. running the audio track through Dragon Naturally Speaking to produce a transcript,
  3. editing the transcript (this is the labor intensive part) to put in punctuation, correct errors, and take out the ums and ahs, and
  4. uploading the transcript into YouTube, which had automated the part about putting in the timings needed to convert the transcript to a file that can be used for captioning.

Here is an example of content produced this way.  The video will be uninteresting content-wise unless you are a student taking intermediate microeconomics.  So 30 seconds should more than suffice.  You may have to push the cc button to get the captions to appear. 




A couple of days ago Drew posted to a campus list for learning technologists that the campus video service, which utilizes Kaltura, has now enabled automated captioning by request.  (The content owner must initiate this.)  Drew duly told us that the quality of the automated captioning will vary depending on a variety of factors.  So there is an expectation that the content owner will edit the video, the way I had done.  There are built in tools in Kaltura for this purpose.

I have my doubts about other faculty members doing such editing, so I wanted to know how good the raw captions are.

I do make screen capture videos now, though not in as great a volume.  The recent ones have not been captioned by me.  I went back to one that I made near the end of the semester last fall.  To my surprise, it had captions.  They are delivered in kind of an eerie way, word by word in a scrolling manner.  Though there is no punctuation, I thought the quality amazingly good.


For the stuff I made back in 2011, I wouldn't have trusted anyone but me to edit the transcript file.  There were some errors that were way off.  In order to correct them you had to know what was right.  But here the errors are really modest, so a student could edit these and I believe do a tolerably good job.

In my class, students who bomb the first exam typically ask if they can do something for extra credit.  I'm not sure of the ethics here, since the extra credit thing should be educative for the student.  But given how scarce instructor time is, if this sort of thing gets the transcript edited, maybe it is the sort of deal with the devil that should happen.

I leave that others.  I will merely note here that that quality of voice to text is getting such that what seemed like an impossible mandate a few years back may be getting do-able now.

No comments: