Grand ideas will sail right past your audience when presented poorly.
This is the first in a series of lessons I’ve learned doing technical and scientific communication with researchers in the video game industry.
In later episodes of this series, we’ll look at videos that reach a general audience. We’ll discover how to shoot a decent-looking interview over Zoom. And still later, we’ll learn about style standards for visual imagery in technical and scientific papers.
But now, we’ll dig into creating video content for a scientific presentation.
What’s the Project?
The example we’ll look at is a game industry research project called Voice2Face. This is a remarkable tool for generating lip sync animation, developed by the research team SEED at Electronic Arts. It’s currently in use in Battlefield, EA Sports FC, and The Sims. Driven by advanced machine learning techniques, it saves animators enormous amounts of tedious work by automatically generating the mouth movements of animated characters based only on recorded speech.
When Voice2Face was still a research project, its breakthrough work was first presented to the public at the Eurographics Symposium on Computer Animation. The video below accompanied the research paper and served as the intro to the conference presentation.
Planning and Script
Visual presentations for technical conferences must balance scientific rigor and showmanship. Everyone likes pretty pictures, but there must be no hint that accuracy has been sacrificed for sex appeal.
Nonetheless, many more people will watch the video than read the paper itself. So, the video must communicate the paper’s results with clarity and brevity, while holding the audience’s interest.
The best place to start is with the intended audience. In this case, the primary audience is the attendees of the SCA conference, who are generally knowledgeable about the field. So we don’t need to worry too much about the use of technical vocabulary. They are expecting to be informed, not necessarily entertained.
Next, we should identify the key takeaways for the video. These can be at least partially derived from the paper’s abstract and conclusion.
However, a video is not a paper, and should not be structured the same way.
Video is a storytelling medium, and even a technical video should have a story arc that runs through it. The Voice2Face video is essentially a product demo with an explanation of the machine learning model in the first half. But other approaches could include:
- Presenting a problem to solve and the solution.
- Comparing before and after states.
- Following a trail of discovery.
- Revealing that a previously held idea turned out to be wrong.
We aimed for a script four to five minutes in length. A good rule of thumb for writing for the spoken word is to assume about 145 words per minute, but that varies by the speaker.
Also, remember that a script is not an essay. Avoid complex sentence structure, which can be difficult to speak aloud and hard for the audience to follow. To make it easy on the audience and the person reading the script, go for one idea per sentence, one sentence per paragraph.
Pro tip for structure: Frontload the script with the most vital conclusions or results. Research papers always present facts first and results afterward. But a video works better by telling your audience upfront where you’re going. Engage people from the first minute, and then backfill the hows and whys.
Visuals
The Voice2Face video made use of materials from the research team’s presentation deck. These were modified to fit the different needs of a video:
- Charts and illustrations were rearranged to fit the video frame and eliminate distracting white space. Smaller elements were removed from the charts to make them quickly readable. Unlike on the printed page, overly complex illustrations will cause the audience’s attention to drift.
- All visuals from the deck were conformed to use a common color scheme. A consistent use of a restricted set of colors produces harmony and cohesiveness.
- Charts and graphs in technical papers are typically presented on white backgrounds, but this can be visually jarring when mixed with images that have dark backgrounds, such as the animated heads in this video. So we put the text and charts on a dark background to provide consistency and a pleasant visual contrast.
- We elected to use minimal transitions and no animations in the charts, in keeping with the dryer tone of a technical presentation.
Note that in the clips of the animated heads, we often blotted out the upper part of the face. This was to keep the viewer’s attention focused on the mouth movements and to avoid the distracting uncanny-valley problem of the model’s rather dead-looking eyes.
Pro tip for color: Check out Viz Palette and Color Brewer for great tools for choosing colors for technical and scientific presentations.
Voiceover
Viewers are usually more tolerant of dodgy video quality than they are of harsh-sounding audio. A crisp and clean voiceover is perhaps the most important single element of any video presentation. Fortunately, there are now AI-driven tools that can help with this.
- Minimize or reduce room echo by placing the microphone close to the speaker. This maximizes clarity of the voice and reduces reverberation.
- Make sure the speaker understands the material. A script full of complex technical jargon can be very difficult to read aloud while communicating the intent behind the words. A lack of familiarity with the material can produce a voiceover with the emphasis in the wrong places or even mispronunciations.
- Use the latest AI-driven tools to clean up the recording. For example, the Adobe Podcast website has an Enhance Speech tool that can make a recording made in your bedroom sound like it was recorded in a studio.
- We used no music for the video, and added the opening bumper only for YouTube.
Next episode: recording great video interviews via Zoom.