Ratios and Rework

This post originally appeared on the Software Carpentry website.

It's been six months and a bit since we started working on Version 4 of this course, so I'd like to share two things we've learned about creating online tutorial videos:

  1. It takes a long time.
  2. A lot of that time feels like it should be unnecessary.

Let's start with "a long time". It takes me 3-5 hours to prepare a good slide deck for an hour-long lecture that I'm going to deliver in person. A slide deck for a screencast takes 8-10, because I need to prepare more slides: since there isn't a lecturer making eye contact and pointing at things to keep viewers engaged, a slide deck for online use has to have many more small transitions.

Writing a script to go with those slides is another 3-5 hours, much of which is spent rearranging the slides as I discover things that don't work as well as I thought. I don't know if it's fair to count this as an extra cost, or whether it's just bringing time I would spend on the second run of a course forward to the first run, but it's still time.

Recording audio takes about 1.5 times as long as the audio (if I stumble, I just pause, take a breath, and restart that sentence), so call it 1.5 hours. Editing takes longer than recording, so let's say 4-5 hours all in (again, including some rework as I notice glitches that escaped me before). Getting everything off my machine and onto software-carpentry.org is another hour by the time all is said and done, so an hour-long lecture made up of 5-6 episodes comes in between 20 and 25 hours.

But now we come to #2: the stuff that feels like it should be unnecessary. If I want to make a change after an episode has been posted—even a small one—it's substantially more work than changing a few slides in a PowerPoint deck would be. First, I need to record new audio, or get whoever originally created the episode to find a microphone and record some snippets of MP3 for me (having the voice change for a slide or two in the middle of a screencast is very jarring). Second, I need to re-export the PowerPoint slides as PNGs, and if the number of slides has changed, rename some of them: Camtasia refers to imported image files by name, and if I simply re-export and re-import, I have to go back and change the time boundaries for all the images after the ones that have been updated.

How bad is this? Well, I just fixed a small mistake in the episode on Python lists. It took almost an hour, start to finish, and the change in audio quality where the fix was made is painful.

And then there's version control. PowerPoint is effectively a binary format: yes, there's XML in there, but version control treats that XML as lines of text, which means its diffs and merges are senseless. I've used HTML and LaTeX in the past—at least I can structure them so that line-oriented diff/merge is sensible—but both formats force a strong separation between text and image, where PowerPoint allows me to mix them freely as I would on a whiteboard. I don't like PowerPoint, but the final result is easier on the eyes than what I can do with today's HTML or LaTeX with equivalent effort.

What do I want instead? I want something that:

  1. Plays nicely with version control (i.e., diff and merge just work).
  2. Allows me to mix text and images freely (PNGs ghettoed in text isn't good enough).
  3. Has stable open source authoring tools with a long likely life in front of them (so that investing in them won't be insanely risky).
  4. Includes a decent text-to-speech engine (something better than Xtranormal's, please) so that when I update what I'm displaying and the script of what I'm saying, I can just push a "compile" button and get a seamless video out the end.

What's that? "And a pony for Christmas?" Yeah, that too. But let me share a secret with you: whoever builds this thing first will be rich, famous, and popular, because this—content creation and maintenance that the average instructor can afford—is the real bottleneck in online education.

Dialogue & Discussion

Comments must follow our Code of Conduct.

Edit this page on Github