Home
technology

Hero Creates AI Doppelgänger of Himself to Get Out of Zoom Video Meetings


Zoom calls with coworkers rises as everyone is self-isolating to stop the spread of COVID-19 (corona virus). In order to skip these lengthy meetings, a technologist has built what he calls a "Zoombot" to double as his doppelgänger. Although the invention is a bit tongue-in-cheek the Zoombot doesn't really blink and uses a robot voice it's a great example of the creative gymnastics that are possible with video conferencing tools.

Zoom has had a hell of a month. First, in the wake of the COVID-19 (corona virus) pandemic, the world collectively embraced the video-conferencing tool because it fosters human connection (hello, virtual happy hours!). But then, employers started bombarding workers with non-stop video meetings, and we rapidly became aware of the myriad security concerns that come with using the app. But every moment in time needs its hero, and that's where Matt Reed comes in. The creative technologist at redpepper, a marketing design firm in Nashville, has introduced an AI-powered "Zoombot" that can sit in on video calls for you.

It all began when Reed noticed someone on Twitter (jokingly) complaining they don't have time to go outside anymore because they're always on Zoom calls. At some level, the project is tongue-in-cheek; Reed isn't a developer by training, and he doesn't claim to be. His AI doppelgänger is a little slow to respond, it doesn't really blink, and it uses a robotic voice similar to voice assistants like Siri or Alexa. But that's the point.

"I didn’t really mean for it to be anything ... but I showed it to a few people at work and they started laughing," Reed says. "I was like, 'This is the kind of thing the world needs right now.'"

To build his bot, Reed recorded himself in Quicktime looking quizzical, confused, opening his mouth, and smiling, and took screenshots. When these images cycle through, it almost looks like Reed has a poor connection and he programmed his bot to say as much.

Then, he built a custom HTML web application that uses an open source library called Artyom.js to listen and respond to prompts. This is what does the brunt of the work, processing audio that comes in through the microphone. Reed had to manually configure it to listen for certain phrases, like "hello," and programmed his application to run commands after hearing certain phrases.

"If it hears 'hello,' it will then start running the hello command, and that will be actual text-to-speech and will trigger that British robot voice to speak that phrase," Reed says. "And it can say anything that you want it to."

Think of Reed's creation like a chatbot or an Alexa voice skill. Chatbots are programmed to understand certain inputs and run commands based on those keywords to give some written response. But both chatbots and Reed's Zoombot are prone to errors due to the natural variability in human language.

Reed speaks as thus; "You have to account for variations. When anyone says 'hello,' for the same command it can listen for 'hello,' 'hi,' and 'how are you doing?'"

Finally, with a piece of software called ManyCam, Reed set up a virtual webcam, using his web app as the source. This creates a video input you can set your Zoom webcam to use, rather than an actual live feed coming through the lens. Reed successfully launched his full prototype after just one morning's worth of work, he says. 

Since Reed sent his Zoombot out into the world, many developers have hopped onto his original idea, perused his GitHub code repository, and made improvements. If you want to take a stab at creating your own Zoombot, head over to Reed's Github account, where the full repository is open source.

Post a Comment