Learning to Reflect Videos

Having revised the "Learning to Reflect" module notes for ANU TechLauncher students, it was time to revise the video. This is a time-consuming and exacting process, even for someone with training in video production.

Last semester I produced this by first making a slideshow presentation, which was also used live in the classroom. I prepared a script based on what was in the module notes, rearranged to match the sequence in the video. I then turned each slide into an image, and the script into synthetic speech. The slides and audio were then imported into a video editing package and timed to match the audio. The results were not perfect, but like Samuel Johnson's piano playing dog: "It's not done well; but you are surprised to find it done at all."  ;-)

This time I decided to try Content Samurai, after it was demonstrated at PitchEd NSW. This is a web based tool which takes a script, and searches for suitable video clips, or still images,  based on the key phrases. The service will then turn the script into synthetic speech timed to the video. This process works remarkably well. The Australian accented male voice provided is much better than those I have used previously. I did have to slow down the narration to 80% of full speed, and lower the volume of the background music to 5%.

After some experimentation I found I could set the system to add a new scene for each paragraph in the script. I could then hide the text, otherwise it would put it as a caption on screen. Also I could upload my slides in place of some of the auto suggested clips. This way my slides are interspersed with the suggested filler scenes.

Here is what the video looks like generated by automatically, with the defaults (video selected from keywords, scene transitions, animated text). This version was not usable:

I then changed the settings, to use only still images, and have one per paragraph, with no animation, or scene transitions:

One problem was that rendering was very slow (but video rendering is always slow). I found it could be faster by using still images, rather than video clips. The rendering took about six minutes for a six minute video. Also I created a plain black template for the slides, with no shaded pattern.

One option I would like is to reduce the bit-rate of the audio. For an educational video you need only low quality mono sound.

One tip is to set your slide maker (I use LibreOffice) for 16:9 format slides, to match a modern widescreen TV. Then generate the slides images at the resolution required. Full HD TV is 1,920x1,080. Content Samurai produced the lower resolution 1,280x720HD TV format typically used for broadcast TV, but down sampled my higher resolution slides very cleanly.

I could not find a way to prepare closed captions within Content Samurai, so I uploaded the video to YouTube, and then downloaded the VTT file it produced. I then uploaded this to the Moodle website, for the students.