# dyncam: a barebones way to replace your webcam feed with a talking avatar

i keep my work laptop under my desk. i plug an external monitor, keyboard, mouse into it. i basically use it as a workstation. therefore its builtin webcam and microphone are not really accessible to me. these days i use a smartphone for the meetings. i was wondering, could i still use my laptop and just present a nice avatar on meetings?

an avatar that would stand still if i wasn't doing active. but it would talk if i talked. or it would be nodding yes or no if i wanted to be nodding. nothing fancy, but streaming simple, manually controlled, precanned animations.

if you are a sensible person, you would use obs studio for this. but obviously i'm too cheapskate to install and learn new free software. so i decided to find my own ways to solve this problem. the information here is absolutely of no use for anyone (if not harmful), i just thought i document it for my own future reference. (so that i have more content to facepalm myself about when i get old.)

i chose phoenix wright from the ace attorney game as my virtual avatar. it's about an attorney who must solve crimes right in the courtroom. his iconic move is saying "objection" whenever he hears a contradiction. it's a gameboy game and i found gif rips online. i found 9 animations i liked:

• 1. just standing (and occasionally blinking)
• 2. normal talking (this is what i'd use when i talk)
• 3. determined talking (he leans in as he talks)
• 4. thinking pose
• 5. talking in the thinking pose
• 6. ashamed talking
• 7. nodding yes
• 8. nodding no
• 9. despair pose

it doesn't have to be anime characters. one could prerecord short loops of themselves too. the point is that i have 9 mpg files called 1.mpg ... 9.mpg, and that all these animations can be looped. i want a little interface where i can choose the currently looping animation.

since the original art was gifs, i converted them into mpg files like this:

  ffmpeg -re -i phoenix/anim008.gif -map 0:v -pix_fmt yuv420p \
    -filter:v crop=960:540:0:0 3.mpg

then i needed a virtual webcam to stream these gifs into. i created one like this:

  modprobe v4l2loopback exclusive_caps=1

now comes the tricky part. it's pretty easy to stream a particular video into such a virtual webcam. however what i want here is a little control box, with which i can dynamically change what i stream into the webcam. one obvious idea is to have ffmpeg loop a video statically into the webcam. whenever i want to change the animation, i kill the current ffmpeg and start a new one with the new animation. this felt unclean. and sometimes it messed up the virtual webcam. it was flaky due to unclean exits, i presume. i needed a saner solution. preferably something that doesn't really require me to code up anything. i found a very ugly hack which allowed me to do this.

ffmpeg has a feature where you can tell it to load a list of files from a file. it can then concatenate those files and loop them. so i created a simple loop.txt containing these two files:

  file a.mpg
  file b.mpg

i instruct ffmpeg to loop concatenating these two files. in order to switch animation, i just replace these files with the new animation. yes, i need two separate files. otherwise ffmpeg doesn't reopen the file. it would keep playing the old animation even if i overwrote the underlying file. i guess this makes sense from an optimization perspective.

so then i just wrote a script that reads digits and rewrites files in a loop. i had one extra requirement though. my animation selection console is running in a terminal. the vidconf software is running in a browser. i wanted to be able to mute and unmute straight from the terminal script. so for that i used xdotool. whenever i press the 0 key i do this:

• 1. focus the browser,
• 2. send the mute/unmute keybinding,
• 3. focus back the terminal.

ugh, i know. for reference here's the full bash monstrosity:

  #!/bin/bash
  echo select the browser window with the vidconf.
  current=$(xdotool getactivewindow)
  browser=$(xdotool selectwindow)
  echo
  
  # remind myself of my options.
  echo '1watch        2talk         3determined'
  echo '4thinking     5thinktalk    6ashamedtalk'
  echo '7yes          8no           9despair'
  
  # set the inital, standing only animation.
  cp 1.mpg a.mpg
  cp 1.mpg b.mpg
  # start the endless ffmpeg stream.
  ffmpeg -loglevel quiet -stream_loop -1 -re -f concat -i loop.txt -map 0:v \
    -f v4l2 -vcodec rawvideo -pix_fmt yuv420p -filter:v crop=960:540:0:0 \
    /dev/video2 &
  # read the requested animation from the keyboard.
  while read -sn1 choice; do
    if ! [[ "$choice" =~ ^[0-9]$ ]]; then
      continue
    fi
    if test "$choice" == 0; then
      # mute or unmute.
      xdotool windowactivate --sync "$browser"
      xdotool key ctrl+d
      xdotool windowactivate "$current" 2>/dev/null
      continue;
    fi
    # set the newly desired animation.
    cp $choice.mpg tmp
    mv tmp a.mpg
    cp $choice.mpg tmp
    mv tmp b.mpg
  done

it's a terrible way to do this but it works okay. i'm pretty sure the gazillion ways to do this better. the main limitation is that it can only switch to the next animation, when the current animation has finished. so it only works with short animations. but that was the case for me so that wasn't a big deal. i don't really use this thing often though, but sounded like a fun enough solution to share.

published on 2020-06-04

to the frontpage